data_source
stringclasses 1
value | prompt
listlengths 2
2
| images
listlengths 0
51
| ability
stringclasses 1
value | reward_model
dict | extra_info
dict |
|---|---|---|---|---|---|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:Blue east hampton pant.Candidates:(1) <image>(2) <image>(3) <image>(4) <image>(5) <image>\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/pants/wide-leg_and_palazzo_pants/91428732/91428732_1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/pants/straight-leg_pants/63390279/63390279_2.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/pants/straight-leg_pants/63390279/63390279_0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/pants/straight-leg_pants/63390279/63390279_1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/pants/straight-leg_pants/63390279/63390279_5.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 2,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/pants/wide-leg_and_palazzo_pants/91428732/91428732_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/pants/straight-leg_pants/63390279/63390279_2.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/pants/straight-leg_pants/63390279/63390279_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/pants/straight-leg_pants/63390279/63390279_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/pants/straight-leg_pants/63390279/63390279_5.jpg"
],
"candidate_texts": [],
"ground_truth_position": 2,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/pants/wide-leg_and_palazzo_pants/91428732/91428732_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/pants/straight-leg_pants/63390279/63390279_2.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/pants/straight-leg_pants/63390279/63390279_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/pants/straight-leg_pants/63390279/63390279_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/pants/straight-leg_pants/63390279/63390279_5.jpg"
],
"num_candidates": 5,
"qid": "1:12752",
"query_image": null,
"query_text": "Blue east hampton pant."
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Candidates:(1) <image>(2) <image>(3) <image>(4) <image>(5) <image>\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/ref/056/086.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/014/901_0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/056/086_1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/063/064_0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/085/030_0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/072/779_0.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 2,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/014/901_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/056/086_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/063/064_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/085/030_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/072/779_0.jpg"
],
"candidate_texts": [],
"ground_truth_position": 2,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/ref/056/086.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/014/901_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/056/086_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/063/064_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/085/030_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/072/779_0.jpg"
],
"num_candidates": 5,
"qid": "4:8936",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/ref/056/086.jpg",
"query_text": ""
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>What is the closest upper taxonomy of this plant?Candidates:(1) Platanus × acerifolia. and comprises a dense spherical cluster of achenes with numerous stiff hairs which aid wind dispersal; the cluster breaks up slowly over the winter to release the numerous 2–3 mm seeds. The London Plane is one of the most efficient trees in removing small particulate pollutants in urban areas.It shares many visual similarities with \"Platanus occidentalis\" (American sycamore), from which it is derived; however, the two species are relatively easy to distinguish, considering the London plane is almost exclusively planted in urban habitats, while \"P. occidentalis\" is most commonly found growing in lowlands and alluvial soils along streams.## Origin.The species.(2) Platanus × acerifolia. exfoliating. The leaves are thick and stiff-textured, broad, palmately lobed, superficially maple-like, the leaf blade 10 - 20 (cm) long and 12 - 25 (cm) broad, with a petiole 3 - 10 (cm) long. The young leaves in spring are coated with minute, fine, stiff hairs at first, but these wear off and by late summer the leaves are hairless or nearly so. The flowers are borne in one to three (most often two) dense spherical inflorescences on a pendulous stem, with male and female flowers on separate stems. The fruit matures in about 6 months, to 2–3 cm diameter,.(3) Platanus × acerifolia. Platanus\" × \"acerifolia, Platanus\" × \"hispanica, or hybrid plane, is a tree in the genus \"Platanus\". It is often known by the synonym London plane, or London planetree. It is usually thought to be a hybrid of \"Platanus orientalis\" (oriental plane) and \"Platanus occidentalis\" (American sycamore). Some authorities think that it may be a cultivar of \"P. orientalis\".## Description.The London plane is a large deciduous tree growing 20 - 30 (m), exceptionally over 40 (m) tall, with a trunk up to 3 (m) or more in circumference. The bark is usually pale grey-green, smooth and exfoliating, or buff-brown and not.(4) Platanus × acerifolia. was formed by hybridization in the 17th century after \"P. orientalis\" and \"P. occidentalis\" had been planted in proximity to one another. It is often said that the hybridization took place in Spain, but it could also have happened in Vauxhall Gardens in London where John Tradescant the Younger discovered the tree in the mid-17th century.The leaf and flower characteristics are intermediate between the two parent species, the leaf being more deeply lobed than \"P. occidentalis\" but less so than \"P. orientalis\", and the seed balls typically two per stem (one in \"P. occidentalis\", 3–6 in \"P. orientalis\"). The hybrid.(5) Platanus × acerifolia. is fertile, and seedlings are occasionally found near mature trees.Controlled reciprocal pollinations between \"P. occidentalis\" and \"P. orientalis\" resulted in good yields of germinable seed and true hybrid seedlings. Crosses of both species, as females, with \"P. racemosa\" and \"P. wrightii\" produced extremely low yields of germinable seed, but true hybrids were obtained from all interspecific combinations. Apomixis (asexual reproduction from non-fertilized seeds) appeared common in \"P. orientalis\".In 1968 and 1970, Frank S. Santamour Jr. recreated the \"P. orientalis\" by \"P. occidentalis\" cross using a \"P. orientalis\" of Turkish origin with American sycamores (\"P. occidentalis\"). The offspring were evaluated following.\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/04/oven_04801471.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 1,
"style": "rule"
}
|
{
"candidate_images": [],
"candidate_texts": [
"Platanus × acerifolia. and comprises a dense spherical cluster of achenes with numerous stiff hairs which aid wind dispersal; the cluster breaks up slowly over the winter to release the numerous 2–3 mm seeds. The London Plane is one of the most efficient trees in removing small particulate pollutants in urban areas.It shares many visual similarities with \"Platanus occidentalis\" (American sycamore), from which it is derived; however, the two species are relatively easy to distinguish, considering the London plane is almost exclusively planted in urban habitats, while \"P. occidentalis\" is most commonly found growing in lowlands and alluvial soils along streams.## Origin.The species.",
"Platanus × acerifolia. exfoliating. The leaves are thick and stiff-textured, broad, palmately lobed, superficially maple-like, the leaf blade 10 - 20 (cm) long and 12 - 25 (cm) broad, with a petiole 3 - 10 (cm) long. The young leaves in spring are coated with minute, fine, stiff hairs at first, but these wear off and by late summer the leaves are hairless or nearly so. The flowers are borne in one to three (most often two) dense spherical inflorescences on a pendulous stem, with male and female flowers on separate stems. The fruit matures in about 6 months, to 2–3 cm diameter,.",
"Platanus × acerifolia. Platanus\" × \"acerifolia, Platanus\" × \"hispanica, or hybrid plane, is a tree in the genus \"Platanus\". It is often known by the synonym London plane, or London planetree. It is usually thought to be a hybrid of \"Platanus orientalis\" (oriental plane) and \"Platanus occidentalis\" (American sycamore). Some authorities think that it may be a cultivar of \"P. orientalis\".## Description.The London plane is a large deciduous tree growing 20 - 30 (m), exceptionally over 40 (m) tall, with a trunk up to 3 (m) or more in circumference. The bark is usually pale grey-green, smooth and exfoliating, or buff-brown and not.",
"Platanus × acerifolia. was formed by hybridization in the 17th century after \"P. orientalis\" and \"P. occidentalis\" had been planted in proximity to one another. It is often said that the hybridization took place in Spain, but it could also have happened in Vauxhall Gardens in London where John Tradescant the Younger discovered the tree in the mid-17th century.The leaf and flower characteristics are intermediate between the two parent species, the leaf being more deeply lobed than \"P. occidentalis\" but less so than \"P. orientalis\", and the seed balls typically two per stem (one in \"P. occidentalis\", 3–6 in \"P. orientalis\"). The hybrid.",
"Platanus × acerifolia. is fertile, and seedlings are occasionally found near mature trees.Controlled reciprocal pollinations between \"P. occidentalis\" and \"P. orientalis\" resulted in good yields of germinable seed and true hybrid seedlings. Crosses of both species, as females, with \"P. racemosa\" and \"P. wrightii\" produced extremely low yields of germinable seed, but true hybrids were obtained from all interspecific combinations. Apomixis (asexual reproduction from non-fertilized seeds) appeared common in \"P. orientalis\".In 1968 and 1970, Frank S. Santamour Jr. recreated the \"P. orientalis\" by \"P. occidentalis\" cross using a \"P. orientalis\" of Turkish origin with American sycamores (\"P. occidentalis\"). The offspring were evaluated following."
],
"ground_truth_position": 1,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/04/oven_04801471.jpg"
],
"num_candidates": 5,
"qid": "6:179511",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/04/oven_04801471.jpg",
"query_text": "What is the closest upper taxonomy of this plant?"
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:Was William P. Rader wearing a necklace and a hat when he carried the Olympic torch in 1996?Candidates:(1) <image>1996 Summer Olympic Torch Relay William P. Rader of Tampa, Florida carrying the torch prior lighting the 4th of July Fireworks celebration.(2) <image>1984 Olympic Torch.(3) <image>POW-MIA remembered through memorial run, Warfit 160915-F-RN654-046 A participant in a 24-hour POW/MIA memorial run carries a flag Sept. 15, 2016, at an outdoor track on Buckley Air Force Base, Colo. Team Buckley members ran in two-hour shifts to remember those service members who sacrificed their lives.(4) In Atlanta in 1996, the cauldron was an artistic scroll decorated in red and gold. It was lit by Muhammad Ali, using a mechanical, self-propelling fuse ball that transported the flame up a wire from the stadium to its cauldron.(5) The route covered 26,875 kilometres (16,699 mi) across the United States and included a trek on the Pony Express, a ride on the Union Pacific Railroad, and a torch was taken into space for the first time. The relay involved over 12,000 torchbearers, including Muhammad Ali, who was chosen to light the Olympic cauldron.\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30108619.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30025401.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30145895.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 1,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30108619.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30025401.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30145895.jpg"
],
"candidate_texts": [
"1996 Summer Olympic Torch Relay William P. Rader of Tampa, Florida carrying the torch prior lighting the 4th of July Fireworks celebration.",
"1984 Olympic Torch.",
"POW-MIA remembered through memorial run, Warfit 160915-F-RN654-046 A participant in a 24-hour POW/MIA memorial run carries a flag Sept. 15, 2016, at an outdoor track on Buckley Air Force Base, Colo. Team Buckley members ran in two-hour shifts to remember those service members who sacrificed their lives.",
"In Atlanta in 1996, the cauldron was an artistic scroll decorated in red and gold. It was lit by Muhammad Ali, using a mechanical, self-propelling fuse ball that transported the flame up a wire from the stadium to its cauldron.",
"The route covered 26,875 kilometres (16,699 mi) across the United States and included a trek on the Pony Express, a ride on the Union Pacific Railroad, and a torch was taken into space for the first time. The relay involved over 12,000 torchbearers, including Muhammad Ali, who was chosen to light the Olympic cauldron."
],
"ground_truth_position": 1,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30108619.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30025401.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30145895.jpg"
],
"num_candidates": 5,
"qid": "2:13430",
"query_image": null,
"query_text": "Was William P. Rader wearing a necklace and a hat when he carried the Olympic torch in 1996?"
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Candidates:(1) <image>(2) <image>(3) <image>(4) <image>(5) <image>\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/ref/038/145.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/057/401_1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/075/194_0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/038/145_0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/014/637_1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/053/213_0.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 3,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/057/401_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/075/194_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/038/145_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/014/637_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/053/213_0.jpg"
],
"candidate_texts": [],
"ground_truth_position": 3,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/ref/038/145.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/057/401_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/075/194_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/038/145_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/014/637_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/053/213_0.jpg"
],
"num_candidates": 5,
"qid": "4:6063",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/ref/038/145.jpg",
"query_text": ""
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Which country does this food come from?Candidates:(1) <image>Soy milk. used as a substitute for dairy milk by individuals who are vegan or are lactose intolerant.Soy milk is also used in making imitation dairy products such as soy yogurt, soy cream, soy kefir and soy-based cheese analogues. It is also used as an ingredient for making milkshakes, pancakes, smoothies, bread, mayonnaise, and baked goods.## Names.In China, the usual term 豆浆 \"dòujiāng\" (\"bean broth\") is used for the traditional watery and beany beverage produced as an intermediate product in the production of tofu, whereas store-bought products designed to imitate the flavor and consistency of dairy milk, may contain a mixture of.(2) <image>Soy milk. Soy milk (simplified Chinese: 豆浆; traditional Chinese: 豆漿) also known as soya milk or soymilk, is a plant-based drink produced by soaking and grinding soybeans, boiling the mixture, and filtering out remaining particulates. It is a stable emulsion of oil, water, and protein. Its original form is an intermediate product of the manufacture of tofu. Originating in China, it became a common beverage in Europe and North America in the latter half of the 20th century, especially as production techniques were developed to give it a taste and consistency more closely resembling that of dairy milk. Soy milk may be.(3) <image>Soy milk. the Eastern Han dynasty unearthed in China, on which is engraved the situation of making soy milk in ancient kitchens.A tofu broth (\"doufujiang\") was used during the Mongol Yuan. As \"doujiang\", this drink remains a common watery form of soy milk in China, usually prepared from fresh soybeans. The compendium of Materia Medica, which was completed in 1578 ad, also has an evaluation of soymilk. Its use increased during the Qing dynasty, apparently due to the discovery that gently heating \"doujiang\" for at least 90 minutes hydrolyzed its raffinose and stachyose, oligosaccharides which can cause flatulence and digestive pain among.(4) <image>Soy milk. lactose-intolerant adults. By the 18th century, it was common enough that street vendors were hawking it; in the 19th, it was also common to take a cup to tofu shops to get hot, fresh \"doujiang\" for breakfast. It was already often paired with youtiao, which was dipped into it. The process was industrialized in early Republican China. By 1929, two Shanghai factories were selling over 1000 bottles a day and another in Beijing was almost as productive itself. Following disruption from the Second World War and the Chinese Civil War, soy milk began to be marketed in soft drink-like fashion.(5) <image>Soy milk. in Hong Kong, Singapore, and Japan in the 1950s.Soymilk was mentioned in various European letters from China beginning in the 17th century. \"Soy milk\" entered the English language (as \"soy-bean milk\") in an 1897 USDA report. Li Yuying established Caséo-Sojaïne, the first soy milk \"dairy\", in Colombes, France, in 1910; he received the first British and American patents for soy milk manufacturing in 1912 and 1913. J.A. Chard began production of \"Soy Lac\" in New York City, United States, in 1917. Harry W. Miller—an American businessman forced to relocate his factory from Shanghai owing to World War II—was similarly compelled.\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/04/oven_04252627.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q192/Q192199.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q192/Q192199.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q192/Q192199.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q192/Q192199.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q192/Q192199.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 1,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q192/Q192199.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q192/Q192199.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q192/Q192199.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q192/Q192199.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q192/Q192199.jpg"
],
"candidate_texts": [
"Soy milk. used as a substitute for dairy milk by individuals who are vegan or are lactose intolerant.Soy milk is also used in making imitation dairy products such as soy yogurt, soy cream, soy kefir and soy-based cheese analogues. It is also used as an ingredient for making milkshakes, pancakes, smoothies, bread, mayonnaise, and baked goods.## Names.In China, the usual term 豆浆 \"dòujiāng\" (\"bean broth\") is used for the traditional watery and beany beverage produced as an intermediate product in the production of tofu, whereas store-bought products designed to imitate the flavor and consistency of dairy milk, may contain a mixture of.",
"Soy milk. Soy milk (simplified Chinese: 豆浆; traditional Chinese: 豆漿) also known as soya milk or soymilk, is a plant-based drink produced by soaking and grinding soybeans, boiling the mixture, and filtering out remaining particulates. It is a stable emulsion of oil, water, and protein. Its original form is an intermediate product of the manufacture of tofu. Originating in China, it became a common beverage in Europe and North America in the latter half of the 20th century, especially as production techniques were developed to give it a taste and consistency more closely resembling that of dairy milk. Soy milk may be.",
"Soy milk. the Eastern Han dynasty unearthed in China, on which is engraved the situation of making soy milk in ancient kitchens.A tofu broth (\"doufujiang\") was used during the Mongol Yuan. As \"doujiang\", this drink remains a common watery form of soy milk in China, usually prepared from fresh soybeans. The compendium of Materia Medica, which was completed in 1578 ad, also has an evaluation of soymilk. Its use increased during the Qing dynasty, apparently due to the discovery that gently heating \"doujiang\" for at least 90 minutes hydrolyzed its raffinose and stachyose, oligosaccharides which can cause flatulence and digestive pain among.",
"Soy milk. lactose-intolerant adults. By the 18th century, it was common enough that street vendors were hawking it; in the 19th, it was also common to take a cup to tofu shops to get hot, fresh \"doujiang\" for breakfast. It was already often paired with youtiao, which was dipped into it. The process was industrialized in early Republican China. By 1929, two Shanghai factories were selling over 1000 bottles a day and another in Beijing was almost as productive itself. Following disruption from the Second World War and the Chinese Civil War, soy milk began to be marketed in soft drink-like fashion.",
"Soy milk. in Hong Kong, Singapore, and Japan in the 1950s.Soymilk was mentioned in various European letters from China beginning in the 17th century. \"Soy milk\" entered the English language (as \"soy-bean milk\") in an 1897 USDA report. Li Yuying established Caséo-Sojaïne, the first soy milk \"dairy\", in Colombes, France, in 1910; he received the first British and American patents for soy milk manufacturing in 1912 and 1913. J.A. Chard began production of \"Soy Lac\" in New York City, United States, in 1917. Harry W. Miller—an American businessman forced to relocate his factory from Shanghai owing to World War II—was similarly compelled."
],
"ground_truth_position": 1,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/04/oven_04252627.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q192/Q192199.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q192/Q192199.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q192/Q192199.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q192/Q192199.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q192/Q192199.jpg"
],
"num_candidates": 5,
"qid": "6:25231",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/04/oven_04252627.jpg",
"query_text": "Which country does this food come from?"
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:DC United s Dwayne De Rosario scored the fastest hat trick in MLS history in a game against Real Salt Lake in 2011.Candidates:(1) <image>(2) <image>(3) <image>(4) <image>(5) <image>\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0650/136.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/washington_post/images/0514/942.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0206/585.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0507/956.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/guardian/images/0416/124.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 3,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0650/136.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/washington_post/images/0514/942.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0206/585.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0507/956.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/guardian/images/0416/124.jpg"
],
"candidate_texts": [],
"ground_truth_position": 3,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0650/136.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/washington_post/images/0514/942.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0206/585.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0507/956.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/guardian/images/0416/124.jpg"
],
"num_candidates": 5,
"qid": "0:171112",
"query_image": null,
"query_text": "DC United s Dwayne De Rosario scored the fastest hat trick in MLS history in a game against Real Salt Lake in 2011."
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Candidates:(1) Black frill dress.(2) Multicolor ruffle dress.(3) Black women's tiered maxi dress.(4) Black ruffled dress.(5) Multicolor lace panel georgette dress.\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/dresses/cocktail_dresses/90981126/90981126_3.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 1,
"style": "rule"
}
|
{
"candidate_images": [],
"candidate_texts": [
"Black frill dress.",
"Multicolor ruffle dress.",
"Black women's tiered maxi dress.",
"Black ruffled dress.",
"Multicolor lace panel georgette dress."
],
"ground_truth_position": 1,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/dresses/cocktail_dresses/90981126/90981126_3.jpg"
],
"num_candidates": 5,
"qid": "1:13585",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/dresses/cocktail_dresses/90981126/90981126_3.jpg",
"query_text": ""
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:A traffic light and cars on a street.Candidates:(1) <image>(2) <image>(3) <image>(4) <image>(5) <image>\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000059106.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000287291.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000154854.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000432273.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000406920.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 2,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000059106.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000287291.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000154854.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000432273.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000406920.jpg"
],
"candidate_texts": [],
"ground_truth_position": 2,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000059106.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000287291.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000154854.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000432273.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000406920.jpg"
],
"num_candidates": 5,
"qid": "9:42735",
"query_image": null,
"query_text": "A traffic light and cars on a street."
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Is a black long sleeved shirt and short-sleeved shirt.Candidates:(1) <image>(2) <image>(3) <image>(4) <image>(5) <image>\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashioniq_images/B00EVALOW8.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashioniq_images/B008B5H91Y.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashioniq_images/B0064I42ZG.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashioniq_images/B005KGF8Y2.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashioniq_images/B00B2JM8PS.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashioniq_images/B004ZC70HA.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 3,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashioniq_images/B008B5H91Y.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashioniq_images/B0064I42ZG.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashioniq_images/B005KGF8Y2.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashioniq_images/B00B2JM8PS.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashioniq_images/B004ZC70HA.jpg"
],
"candidate_texts": [],
"ground_truth_position": 3,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashioniq_images/B00EVALOW8.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashioniq_images/B008B5H91Y.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashioniq_images/B0064I42ZG.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashioniq_images/B005KGF8Y2.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashioniq_images/B00B2JM8PS.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashioniq_images/B004ZC70HA.jpg"
],
"num_candidates": 5,
"qid": "7:10060",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashioniq_images/B00EVALOW8.jpg",
"query_text": "Is a black long sleeved shirt and short-sleeved shirt."
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:The metal with the symbol Cs was discovered in water from where?Candidates:(1) Potassium is a chemical element with the symbol K (from Neo-Latin kalium) and atomic number 19. Potassium is a silvery-white metal that is soft enough to be cut with a knife with little force. Potassium metal reacts rapidly with atmospheric oxygen to form flaky white potassium peroxide in only seconds of exposure.(2) After reduction of the hexachloroplatinate with hydrogen, caesium and rubidium were separated by the difference in solubility of their carbonates in alcohol. The process yielded 9.2 grams (0.32 oz) of rubidium chloride and 7.3 grams (0.26 oz) of caesium chloride from the initial 44,000 litres of mineral water.(3) Bismuth is a chemical element with the symbol Bi and atomic number 83. It is a pentavalent post-transition metal and one of the pnictogens with chemical properties resembling its lighter group 15 siblings arsenic and antimony. Elemental bismuth may occur naturally, although its sulfide and oxide form important commercial ores.(4) Caesium ( IUPAC spelling) ( also spelled cesium in American English) is a chemical element with the symbol Cs and atomic number 55. It is a soft, silvery-golden alkali metal with a melting point of 28.5 °C (83.3 °F), which makes it one of only five elemental metals that are liquid at or near room temperature.(5) In 1860, Robert Bunsen and Gustav Kirchhoff discovered caesium in the mineral water from Dürkheim, Germany. Because of the bright blue lines in the emission spectrum, they derived the name from the Latin word caesius, meaning sky-blue.[note 6][66][67][68] Caesium was the first element to be discovered with a spectroscope, which had been invented by Bunsen and Kirchhoff only a year previously.[15].\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[] |
multimodal_reranking
|
{
"ground_truth": 4,
"style": "rule"
}
|
{
"candidate_images": [],
"candidate_texts": [
"Potassium is a chemical element with the symbol K (from Neo-Latin kalium) and atomic number 19. Potassium is a silvery-white metal that is soft enough to be cut with a knife with little force. Potassium metal reacts rapidly with atmospheric oxygen to form flaky white potassium peroxide in only seconds of exposure.",
"After reduction of the hexachloroplatinate with hydrogen, caesium and rubidium were separated by the difference in solubility of their carbonates in alcohol. The process yielded 9.2 grams (0.32 oz) of rubidium chloride and 7.3 grams (0.26 oz) of caesium chloride from the initial 44,000 litres of mineral water.",
"Bismuth is a chemical element with the symbol Bi and atomic number 83. It is a pentavalent post-transition metal and one of the pnictogens with chemical properties resembling its lighter group 15 siblings arsenic and antimony. Elemental bismuth may occur naturally, although its sulfide and oxide form important commercial ores.",
"Caesium ( IUPAC spelling) ( also spelled cesium in American English) is a chemical element with the symbol Cs and atomic number 55. It is a soft, silvery-golden alkali metal with a melting point of 28.5 °C (83.3 °F), which makes it one of only five elemental metals that are liquid at or near room temperature.",
"In 1860, Robert Bunsen and Gustav Kirchhoff discovered caesium in the mineral water from Dürkheim, Germany. Because of the bright blue lines in the emission spectrum, they derived the name from the Latin word caesius, meaning sky-blue.[note 6][66][67][68] Caesium was the first element to be discovered with a spectroscope, which had been invented by Bunsen and Kirchhoff only a year previously.[15]."
],
"ground_truth_position": 4,
"images": [],
"num_candidates": 5,
"qid": "2:28426",
"query_image": null,
"query_text": "The metal with the symbol Cs was discovered in water from where?"
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Candidates:(1) A little white lamb looking through a fence.(2) A sheep laying in a field behind a fence.(3) An animal sits in the grass by the fence.(4) A close up of a sheep laying in a field behind a fence.(5) A white sheep sitting in grass next to a wire fence.\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000022578.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 1,
"style": "rule"
}
|
{
"candidate_images": [],
"candidate_texts": [
"A little white lamb looking through a fence.",
"A sheep laying in a field behind a fence.",
"An animal sits in the grass by the fence.",
"A close up of a sheep laying in a field behind a fence.",
"A white sheep sitting in grass next to a wire fence."
],
"ground_truth_position": 1,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000022578.jpg"
],
"num_candidates": 5,
"qid": "9:205461",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000022578.jpg",
"query_text": ""
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:Who are the champions.Candidates:(1) <image>(2) <image>(3) <image>(4) <image>(5) <image>\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/guardian/images/0465/872.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/guardian/images/0457/770.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/guardian/images/0112/349.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/guardian/images/0583/369.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/guardian/images/0230/776.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 4,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/guardian/images/0465/872.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/guardian/images/0457/770.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/guardian/images/0112/349.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/guardian/images/0583/369.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/guardian/images/0230/776.jpg"
],
"candidate_texts": [],
"ground_truth_position": 4,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/guardian/images/0465/872.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/guardian/images/0457/770.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/guardian/images/0112/349.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/guardian/images/0583/369.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/guardian/images/0230/776.jpg"
],
"num_candidates": 5,
"qid": "0:58321",
"query_image": null,
"query_text": "Who are the champions."
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Candidates:(1) A batter has just hit the ball at home plate.(2) A baseball player holding a bat next to home plate.(3) A member of the Giants swinging the bat during the baseball game.(4) A baseball playing swinging a baseball bat.(5) The man is hitting the baseball during the game.\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000485179.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 1,
"style": "rule"
}
|
{
"candidate_images": [],
"candidate_texts": [
"A batter has just hit the ball at home plate.",
"A baseball player holding a bat next to home plate.",
"A member of the Giants swinging the bat during the baseball game.",
"A baseball playing swinging a baseball bat.",
"The man is hitting the baseball during the game."
],
"ground_truth_position": 1,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000485179.jpg"
],
"num_candidates": 5,
"qid": "9:34252",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000485179.jpg",
"query_text": ""
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:A keyboard and mouse along with other items on the desk.Candidates:(1) <image>(2) <image>(3) <image>(4) <image>(5) <image>\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000197600.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000247921.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000567686.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000572289.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000480015.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 3,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000197600.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000247921.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000567686.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000572289.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000480015.jpg"
],
"candidate_texts": [],
"ground_truth_position": 3,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000197600.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000247921.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000567686.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000572289.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000480015.jpg"
],
"num_candidates": 5,
"qid": "9:1808",
"query_image": null,
"query_text": "A keyboard and mouse along with other items on the desk."
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:Did John Wesley and Sarah Thomas star in shows broadcast on the same or different continents?Candidates:(1) This in turn led to his performing a number of Coward's monologues and songs on stage, as part of the \"Hollywood Party\" sequence in his one-man show Star Struck which premiered at the Fringe in 2003, built around the Coward monologue Mad Dogs and Englishmen.(2) Catriona MacColl stars as the eponymous Oscar François de Jarjayes, with Barry Stokes as her lover André Grandier, alongside Jonas Bergström, Christine Böhm, Mark Kingston, Georges Wilson, Sue Lloyd, Martin Potter, and Anouska Hempel. A young Lambert Wilson appears in a minor role.(3) John Wesley (August 3, 1947 – September 7, 2019) was an American actor who worked in many television series, including The Fresh Prince of Bel Air, and in Superhuman Samurai Syber-Squad, where he played the role of Principal Pratchert. He also had roles in many feature films as well as theater productions.(4) In 1985 she joined the cast of the internationally successful BBC television sitcom Last Of The Summer Wine as Glenda, the daughter of Thora Hird 's character. She stayed with the series for 25 years, until it came to an end in 2010. Her most recent television guest star roles were in episodes of Heartbeat, The Bill and Doctors .(5) In 1971, Elliman moved to New York City for the Broadway production of Jesus Christ, Superstar, where she met her first husband, Bill Oakes, who was President of Robert Stigwood's RSO Records.\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[] |
multimodal_reranking
|
{
"ground_truth": 3,
"style": "rule"
}
|
{
"candidate_images": [],
"candidate_texts": [
"This in turn led to his performing a number of Coward's monologues and songs on stage, as part of the \"Hollywood Party\" sequence in his one-man show Star Struck which premiered at the Fringe in 2003, built around the Coward monologue Mad Dogs and Englishmen.",
"Catriona MacColl stars as the eponymous Oscar François de Jarjayes, with Barry Stokes as her lover André Grandier, alongside Jonas Bergström, Christine Böhm, Mark Kingston, Georges Wilson, Sue Lloyd, Martin Potter, and Anouska Hempel. A young Lambert Wilson appears in a minor role.",
"John Wesley (August 3, 1947 – September 7, 2019) was an American actor who worked in many television series, including The Fresh Prince of Bel Air, and in Superhuman Samurai Syber-Squad, where he played the role of Principal Pratchert. He also had roles in many feature films as well as theater productions.",
"In 1985 she joined the cast of the internationally successful BBC television sitcom Last Of The Summer Wine as Glenda, the daughter of Thora Hird 's character. She stayed with the series for 25 years, until it came to an end in 2010. Her most recent television guest star roles were in episodes of Heartbeat, The Bill and Doctors .",
"In 1971, Elliman moved to New York City for the Broadway production of Jesus Christ, Superstar, where she met her first husband, Bill Oakes, who was President of Robert Stigwood's RSO Records."
],
"ground_truth_position": 3,
"images": [],
"num_candidates": 5,
"qid": "2:19654",
"query_image": null,
"query_text": "Did John Wesley and Sarah Thomas star in shows broadcast on the same or different continents?"
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Candidates:(1) A sepia photo at a rail yard with people standing on the tracks.(2) People are crowding around on the tracks in front of an old fashioned train.(3) A photo from 1897 of a railroad crowded by people.(4) A couple of people are walking near an old train.(5) An old photo of a train in a station with a crowd.\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000271806.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 1,
"style": "rule"
}
|
{
"candidate_images": [],
"candidate_texts": [
"A sepia photo at a rail yard with people standing on the tracks.",
"People are crowding around on the tracks in front of an old fashioned train.",
"A photo from 1897 of a railroad crowded by people.",
"A couple of people are walking near an old train.",
"An old photo of a train in a station with a crowd."
],
"ground_truth_position": 1,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000271806.jpg"
],
"num_candidates": 5,
"qid": "9:43459",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000271806.jpg",
"query_text": ""
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Many monkeys walking in the road.Candidates:(1) <image>(2) <image>(3) <image>(4) <image>(5) <image>\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/96/train-2880-3-img1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/73/train-6399-3-img0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/31/train-10029-3-img0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/22/train-3894-2-img1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/66/train-3090-2-img0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/78/train-10145-2-img0.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 1,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/73/train-6399-3-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/31/train-10029-3-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/22/train-3894-2-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/66/train-3090-2-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/78/train-10145-2-img0.jpg"
],
"candidate_texts": [],
"ground_truth_position": 1,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/96/train-2880-3-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/73/train-6399-3-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/31/train-10029-3-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/22/train-3894-2-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/66/train-3090-2-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/78/train-10145-2-img0.jpg"
],
"num_candidates": 5,
"qid": "8:16433",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/96/train-2880-3-img1.jpg",
"query_text": "Many monkeys walking in the road."
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Which species of bird is this?Candidates:(1) Plain chachalaca. The plain chachalaca (\"Ortalis vetula\") is a large bird in the chachalaca, guan and curassow family Cracidae. It breeds in tropical and subtropical environments from mezquital thickets in the Rio Grande Valley in southernmost Texas, United States to northernmost Costa Rica. In Central America, this species occurs in the Pacific lowlands from Chiapas, Mexico to northern Nicaragua and as a separate population in Costa Rica, where its range is separated by a short distance, as a disjunct population.This species frequents dry and moist forests, especially where interspersed with scrub and savanna. Usually found in groups of up to 15 birds,.(2) Plain chachalaca. the plain chachalaca is furtive and wary and prefers to escape from danger by running swiftly on the ground or leaping and gliding through brushy tangles. The plain chachalaca feeds in trees or on the ground on fruit (figs, palms, Sapotaceae), seeds, leaves, and flowers. It is sometimes a pest of crops such as tomatoes and cucumbers.The call is a loud, raucous \"RAW-pa-haw\" or \"cha-cha-LAW-ka\", often by several birds in a rhythmical chorus, especially in early morning and evening, usually from well up in trees. It also produces peeping whistles and cackles. Others describe chachalaca calls as irritating noises mimicking.(3) Plain chachalaca. The plain chachalaca (\"Ortalis vetula\") is a large bird in the chachalaca, guan and curassow family Cracidae. It breeds in tropical and subtropical environments from mezquital thickets in the Rio Grande Valley in southernmost Texas, United States to northernmost Costa Rica. In Central America, this species occurs in the Pacific lowlands from Chiapas, Mexico to northern Nicaragua and as a separate population in Costa Rica, where its range is separated by a short distance, as a disjunct population.This species frequents dry and moist forests, especially where interspersed with scrub and savanna. Usually found in groups of up to 15 birds,.(4) Plain chachalaca. to the little chachalaca than the plain chachalaca.There are five recognized subspecies: \"O. v. deschauenseei\" – - Utila chachalaca: Utila Island, off Honduras \"O. v. mccallii\" – : extreme southern Texas in the United States to northern Veracruz in Mexico \"O. v. intermedia\" – : southern Yucatán (Mexico) to Belize and Guatemala, (sometime included in \"O. v. vetula\") \"O. v. pallidiventris\" – : northern Yucatán Peninsula \"O. v. vetula\" – : nominate, southern Veracruz in Mexico to northwest Costa RicaThe isolated population in Costa Rica is assigned to the nominate race of the plain chachalaca, but has also been assigned.(5) Plain chachalaca. a bunch of men arguing. The plain chachalaca typically breeds in the early wet season. The nest is a shallow saucer of twigs and plant fibers, lined with leaves, in thick vegetation. The clutch is 2–4 rough-shelled white to cream eggs.## Taxonomy.The plain chachalaca is related to the West Mexican chachalaca and the white-bellied chachalaca, and both of these were formerly considered a subspecies of the plain chachalaca . Today all major authorities recognize them as separate species. Their ranges slightly overlap in Mexico, but hybrids have not been reported between them. The white-bellied chachalaca may be more closely related.\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/00/oven_00773981.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 1,
"style": "rule"
}
|
{
"candidate_images": [],
"candidate_texts": [
"Plain chachalaca. The plain chachalaca (\"Ortalis vetula\") is a large bird in the chachalaca, guan and curassow family Cracidae. It breeds in tropical and subtropical environments from mezquital thickets in the Rio Grande Valley in southernmost Texas, United States to northernmost Costa Rica. In Central America, this species occurs in the Pacific lowlands from Chiapas, Mexico to northern Nicaragua and as a separate population in Costa Rica, where its range is separated by a short distance, as a disjunct population.This species frequents dry and moist forests, especially where interspersed with scrub and savanna. Usually found in groups of up to 15 birds,.",
"Plain chachalaca. the plain chachalaca is furtive and wary and prefers to escape from danger by running swiftly on the ground or leaping and gliding through brushy tangles. The plain chachalaca feeds in trees or on the ground on fruit (figs, palms, Sapotaceae), seeds, leaves, and flowers. It is sometimes a pest of crops such as tomatoes and cucumbers.The call is a loud, raucous \"RAW-pa-haw\" or \"cha-cha-LAW-ka\", often by several birds in a rhythmical chorus, especially in early morning and evening, usually from well up in trees. It also produces peeping whistles and cackles. Others describe chachalaca calls as irritating noises mimicking.",
"Plain chachalaca. The plain chachalaca (\"Ortalis vetula\") is a large bird in the chachalaca, guan and curassow family Cracidae. It breeds in tropical and subtropical environments from mezquital thickets in the Rio Grande Valley in southernmost Texas, United States to northernmost Costa Rica. In Central America, this species occurs in the Pacific lowlands from Chiapas, Mexico to northern Nicaragua and as a separate population in Costa Rica, where its range is separated by a short distance, as a disjunct population.This species frequents dry and moist forests, especially where interspersed with scrub and savanna. Usually found in groups of up to 15 birds,.",
"Plain chachalaca. to the little chachalaca than the plain chachalaca.There are five recognized subspecies: \"O. v. deschauenseei\" – - Utila chachalaca: Utila Island, off Honduras \"O. v. mccallii\" – : extreme southern Texas in the United States to northern Veracruz in Mexico \"O. v. intermedia\" – : southern Yucatán (Mexico) to Belize and Guatemala, (sometime included in \"O. v. vetula\") \"O. v. pallidiventris\" – : northern Yucatán Peninsula \"O. v. vetula\" – : nominate, southern Veracruz in Mexico to northwest Costa RicaThe isolated population in Costa Rica is assigned to the nominate race of the plain chachalaca, but has also been assigned.",
"Plain chachalaca. a bunch of men arguing. The plain chachalaca typically breeds in the early wet season. The nest is a shallow saucer of twigs and plant fibers, lined with leaves, in thick vegetation. The clutch is 2–4 rough-shelled white to cream eggs.## Taxonomy.The plain chachalaca is related to the West Mexican chachalaca and the white-bellied chachalaca, and both of these were formerly considered a subspecies of the plain chachalaca . Today all major authorities recognize them as separate species. Their ranges slightly overlap in Mexico, but hybrids have not been reported between them. The white-bellied chachalaca may be more closely related."
],
"ground_truth_position": 1,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/00/oven_00773981.jpg"
],
"num_candidates": 5,
"qid": "5:137206",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/00/oven_00773981.jpg",
"query_text": "Which species of bird is this?"
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>What is shown in the photo?Candidates:(1) Red-backed salamander. form. Additionally, lead-backed morphs withdraw from surface activity earlier in the autumn than red-backed morphs, presumably to avoid cooling temperatures. Standard metabolic rate has also been found to differ between the morphs at certain temperatures, with significantly lower metabolic rates being displayed by the lead-backed form at 15 °C; in the same study, lead-backed individuals were also more active on the ground surface at this temperature. These findings suggest that the lead-backed color variant is less tolerant of cool temperatures than the red-backed color variant, and that the two color forms differ physiologically and behaviorally at certain temperatures.An alternate explanation.(2) African clawed frog. as a more viable model for genetics.Roger Wolcott Sperry used \"X. laevis\" for his famous experiments describing the development of the visual system. These experiments led to the formulation of the chemoaffinity hypothesis.\"Xenopus\" oocytes provide an important expression system for molecular biology. By injecting DNA or mRNA into the oocyte or developing embryo, scientists can study the protein products in a controlled system. This allows rapid functional expression of manipulated DNAs (or mRNA). This is particularly useful in electrophysiology, where the ease of recording from the oocyte makes expression of membrane channels attractive. One challenge of oocyte work is eliminating.(3) Canyon tree frog. are lighter colored, and those from regions composed mostly of granite can even be pink in coloration. Most have bright yellow in their groin regions, and faded banding on their legs.## Behavior.Canyon tree frogs are mostly nocturnal and carnivorous. They are typically found in semiarid, rocky habitats near a permanent water source. Breeding occurs during the spring rains, and large, floating egg masses of 100 or more eggs are laid on the water. During periods of low rainfall, the frogs will take refuge in rock/bolder crevices.Desert Frogs of Tucson, AZReptiles of Arizona Hyla arenicolor movies Canyon Treefrog audio recordings.(4) Canyon tree frog. The canyon tree frog (\"Dryophytes arenicolor\") is a species of tree frog native to the rocky plateau areas of southern United States, primarily in New Mexico and Arizona, but it also ranges to Utah, Texas, and Colorado, and as far south as the Mexican states of Michoacán, México, Guanajuato, Guerrero, and Oaxaca.## Description.Canyon tree frogs grow to 5.0 - 5.5 (cm) in length, and are typically brown, grey-brown, or grey-green in color, often with darker-colored blotching. They can vary considerably, but usually match the soil or rock coloration of their native habitats to serve as camouflage. Those from limestone habitats.(5) Canyon tree frog. The canyon tree frog (\"Dryophytes arenicolor\") is a species of tree frog native to the rocky plateau areas of southern United States, primarily in New Mexico and Arizona, but it also ranges to Utah, Texas, and Colorado, and as far south as the Mexican states of Michoacán, México, Guanajuato, Guerrero, and Oaxaca.## Description.Canyon tree frogs grow to 5.0 - 5.5 (cm) in length, and are typically brown, grey-brown, or grey-green in color, often with darker-colored blotching. They can vary considerably, but usually match the soil or rock coloration of their native habitats to serve as camouflage. Those from limestone habitats.\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/00/oven_00595919.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 3,
"style": "rule"
}
|
{
"candidate_images": [],
"candidate_texts": [
"Red-backed salamander. form. Additionally, lead-backed morphs withdraw from surface activity earlier in the autumn than red-backed morphs, presumably to avoid cooling temperatures. Standard metabolic rate has also been found to differ between the morphs at certain temperatures, with significantly lower metabolic rates being displayed by the lead-backed form at 15 °C; in the same study, lead-backed individuals were also more active on the ground surface at this temperature. These findings suggest that the lead-backed color variant is less tolerant of cool temperatures than the red-backed color variant, and that the two color forms differ physiologically and behaviorally at certain temperatures.An alternate explanation.",
"African clawed frog. as a more viable model for genetics.Roger Wolcott Sperry used \"X. laevis\" for his famous experiments describing the development of the visual system. These experiments led to the formulation of the chemoaffinity hypothesis.\"Xenopus\" oocytes provide an important expression system for molecular biology. By injecting DNA or mRNA into the oocyte or developing embryo, scientists can study the protein products in a controlled system. This allows rapid functional expression of manipulated DNAs (or mRNA). This is particularly useful in electrophysiology, where the ease of recording from the oocyte makes expression of membrane channels attractive. One challenge of oocyte work is eliminating.",
"Canyon tree frog. are lighter colored, and those from regions composed mostly of granite can even be pink in coloration. Most have bright yellow in their groin regions, and faded banding on their legs.## Behavior.Canyon tree frogs are mostly nocturnal and carnivorous. They are typically found in semiarid, rocky habitats near a permanent water source. Breeding occurs during the spring rains, and large, floating egg masses of 100 or more eggs are laid on the water. During periods of low rainfall, the frogs will take refuge in rock/bolder crevices.Desert Frogs of Tucson, AZReptiles of Arizona Hyla arenicolor movies Canyon Treefrog audio recordings.",
"Canyon tree frog. The canyon tree frog (\"Dryophytes arenicolor\") is a species of tree frog native to the rocky plateau areas of southern United States, primarily in New Mexico and Arizona, but it also ranges to Utah, Texas, and Colorado, and as far south as the Mexican states of Michoacán, México, Guanajuato, Guerrero, and Oaxaca.## Description.Canyon tree frogs grow to 5.0 - 5.5 (cm) in length, and are typically brown, grey-brown, or grey-green in color, often with darker-colored blotching. They can vary considerably, but usually match the soil or rock coloration of their native habitats to serve as camouflage. Those from limestone habitats.",
"Canyon tree frog. The canyon tree frog (\"Dryophytes arenicolor\") is a species of tree frog native to the rocky plateau areas of southern United States, primarily in New Mexico and Arizona, but it also ranges to Utah, Texas, and Colorado, and as far south as the Mexican states of Michoacán, México, Guanajuato, Guerrero, and Oaxaca.## Description.Canyon tree frogs grow to 5.0 - 5.5 (cm) in length, and are typically brown, grey-brown, or grey-green in color, often with darker-colored blotching. They can vary considerably, but usually match the soil or rock coloration of their native habitats to serve as camouflage. Those from limestone habitats."
],
"ground_truth_position": 3,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/00/oven_00595919.jpg"
],
"num_candidates": 5,
"qid": "5:120746",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/00/oven_00595919.jpg",
"query_text": "What is shown in the photo?"
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Candidates:(1) A skier in the snow heading into a wooded area.(2) A person riding skis across a snow covered forest.(3) A snowy woodland with a person on skis exploring it.(4) A woman cross country skiing in the woods.(5) A skier is going cross country through the woods.\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000441929.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 1,
"style": "rule"
}
|
{
"candidate_images": [],
"candidate_texts": [
"A skier in the snow heading into a wooded area.",
"A person riding skis across a snow covered forest.",
"A snowy woodland with a person on skis exploring it.",
"A woman cross country skiing in the woods.",
"A skier is going cross country through the woods."
],
"ground_truth_position": 1,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000441929.jpg"
],
"num_candidates": 5,
"qid": "9:37816",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000441929.jpg",
"query_text": ""
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:White seamed tank dress w/ lace trim.Candidates:(1) <image>(2) <image>(3) <image>(4) <image>(5) <image>\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/dresses/casual_and_day_dresses/89570674/89570674_0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/dresses/casual_and_day_dresses/89570674/89570674_1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/dresses/casual_and_day_dresses/89570674/89570674_2.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/dresses/mini_and_short_dresses/91429192/91429192_0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/dresses/casual_and_day_dresses/89570674/89570674_3.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 1,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/dresses/casual_and_day_dresses/89570674/89570674_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/dresses/casual_and_day_dresses/89570674/89570674_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/dresses/casual_and_day_dresses/89570674/89570674_2.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/dresses/mini_and_short_dresses/91429192/91429192_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/dresses/casual_and_day_dresses/89570674/89570674_3.jpg"
],
"candidate_texts": [],
"ground_truth_position": 1,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/dresses/casual_and_day_dresses/89570674/89570674_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/dresses/casual_and_day_dresses/89570674/89570674_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/dresses/casual_and_day_dresses/89570674/89570674_2.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/dresses/mini_and_short_dresses/91429192/91429192_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/dresses/casual_and_day_dresses/89570674/89570674_3.jpg"
],
"num_candidates": 5,
"qid": "1:23768",
"query_image": null,
"query_text": "White seamed tank dress w/ lace trim."
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:Do the flowers of the Red hot pokers blossom in only one direction?Candidates:(1) <image>'Crocosmia × crocosmiiflora' Montbretia in Walled Garden border of Parham House, West Sussex, England 3 Crocosmia, commonly \"Montbretia\", with a selection of flowers and shrubs in a border of the Walled Garden of Parham House, in West Sussex, England.(2) <image>Potentilla fruticosa Red Ace 01 Potentilla fruticosa'Red Ace'.(3) <image>Jonge scheuten van Euphorbia griffithii 'Fireglow'. Locatie, Tuinreservaat Jonker vallei Young shoots of Euphorbia griffithii 'Fire Glow'. Location: Garden reservaat Valley Jonker.(4) <image>Kniphofia uvaria5.(5) <image>Gardenology.org-IMG 1172 bbg09 Photographed at Brooklyn Botanic Garden (New York) in September This photo is from Gardenology.org and is available under CC-BY-SA 3.0 license. If you use it, please include attribution to Gardenology.org, with a link if the media allows it. If the name is missing, click here.\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30188042.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30172484.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30369463.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30237006.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30291791.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 4,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30188042.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30172484.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30369463.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30237006.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30291791.jpg"
],
"candidate_texts": [
"'Crocosmia × crocosmiiflora' Montbretia in Walled Garden border of Parham House, West Sussex, England 3 Crocosmia, commonly \"Montbretia\", with a selection of flowers and shrubs in a border of the Walled Garden of Parham House, in West Sussex, England.",
"Potentilla fruticosa Red Ace 01 Potentilla fruticosa'Red Ace'.",
"Jonge scheuten van Euphorbia griffithii 'Fireglow'. Locatie, Tuinreservaat Jonker vallei Young shoots of Euphorbia griffithii 'Fire Glow'. Location: Garden reservaat Valley Jonker.",
"Kniphofia uvaria5.",
"Gardenology.org-IMG 1172 bbg09 Photographed at Brooklyn Botanic Garden (New York) in September This photo is from Gardenology.org and is available under CC-BY-SA 3.0 license. If you use it, please include attribution to Gardenology.org, with a link if the media allows it. If the name is missing, click here."
],
"ground_truth_position": 4,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30188042.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30172484.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30369463.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30237006.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30291791.jpg"
],
"num_candidates": 5,
"qid": "2:5915",
"query_image": null,
"query_text": "Do the flowers of the Red hot pokers blossom in only one direction?"
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Lilac vase on a brown table instead of a white vase with birds on a white background.Candidates:(1) <image>(2) <image>(3) <image>(4) <image>(5) <image>\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/62/train-1528-1-img0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/69/train-9058-2-img0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/77/train-11629-0-img0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/96/train-12508-1-img1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/3/train-7263-0-img0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/69/train-12742-1-img1.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 4,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/69/train-9058-2-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/77/train-11629-0-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/96/train-12508-1-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/3/train-7263-0-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/69/train-12742-1-img1.jpg"
],
"candidate_texts": [],
"ground_truth_position": 4,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/62/train-1528-1-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/69/train-9058-2-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/77/train-11629-0-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/96/train-12508-1-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/3/train-7263-0-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/69/train-12742-1-img1.jpg"
],
"num_candidates": 5,
"qid": "8:12679",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/62/train-1528-1-img0.jpg",
"query_text": "Lilac vase on a brown table instead of a white vase with birds on a white background."
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Candidates:(1) Fishnet white top.(2) Black mesh top.(3) Blue fluted lace-hem jersey top.(4) Gray mesh overlay tunic.(5) Blue long sleeve top.\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/tops/short_sleeve_tops/90985872/90985872_2.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 4,
"style": "rule"
}
|
{
"candidate_images": [],
"candidate_texts": [
"Fishnet white top.",
"Black mesh top.",
"Blue fluted lace-hem jersey top.",
"Gray mesh overlay tunic.",
"Blue long sleeve top."
],
"ground_truth_position": 4,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/tops/short_sleeve_tops/90985872/90985872_2.jpg"
],
"num_candidates": 5,
"qid": "1:26258",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/tops/short_sleeve_tops/90985872/90985872_2.jpg",
"query_text": ""
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Candidates:(1) A man with a toothbrush in his mouth and uncombed hair takes a picture of himself at his computer desk.(2) A man has a blue toothbrush in his mouth.(3) A man standing by a desk with a toothbrush in his mouth.(4) A man brushing his teeth with a tooth brush.(5) A man is brushing his teeth near a laptop.\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000536417.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 1,
"style": "rule"
}
|
{
"candidate_images": [],
"candidate_texts": [
"A man with a toothbrush in his mouth and uncombed hair takes a picture of himself at his computer desk.",
"A man has a blue toothbrush in his mouth.",
"A man standing by a desk with a toothbrush in his mouth.",
"A man brushing his teeth with a tooth brush.",
"A man is brushing his teeth near a laptop."
],
"ground_truth_position": 1,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000536417.jpg"
],
"num_candidates": 5,
"qid": "9:165338",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000536417.jpg",
"query_text": ""
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>What is the brand of this vehicle?Candidates:(1) <image>Hummer H2. Air compressor with road assistance kit, and 20\" chrome wheels (slightly different from stock wheels).## 2008 updates.For 2008, the Hummer H2 and H2 SUT received an update. While largely unchanged on the exterior, the H2 and H2 SUT interiors were redesigned. This included new instrument cluster with improved gauges and Driver Information Center (DIC), a new leather-wrapped steering wheel, three new radios with a Bose premium audio system and auxiliary audio inputs (including a new touchscreen GPS navigation radio with DVD audio and video playback while in park, and XM Nav-Traffic capabilities), a new rear seat DVD entertainment system, Bluetooth.(2) <image>Hummer H2. The Hummer H2 is a large SUV that was marketed by Hummer and built in the AM General facility under contract from General Motors from 2002 to 2009. It is based on a modified GMT820 GM three-quarter-ton pickup truck in the front and a half-ton 1500 frame in the rear. A four-door pickup truck version with a midgate that opens the vehicle's interior to the external cargo bed was introduced for 2005 as the H2 SUT (sport utility truck).## Background.In 2001 and 2002, GM allowed journalists to examine early versions of the Hummer H2, which were still under development at.(3) <image>Hummer H2. seats, dual memory system, BOSE premium sound system, single-CD/cassette player and later in 2004, a 6-disc CD changer, then in 2008,a single CD-player with MP3 capability, an auxiliary input jack and DVD player, outside-temperature indicator, compass, rear radio controls, independent front torsion bar suspension, rear 5-link coil spring suspension specially created for the H2, oversized tires with HUMMER wheels, universal garage door opener and remote engine start (2008–2009).## Optional.Options for the H2 include adjustable rear suspension [which is included within the Adventure Package], a wide power sunroof, rearview camera, DVD entertainment system, navigation system, ladder, custom grilles, side step bars,.(4) <image>Hummer H2. hands-free calling capabilities, enhanced voice activation, new OnStar hardware with buttons moved from the rearview mirror to the overhead console, a lower dashboard-mounted radio, new dual-zone climate controls, new rear seat audio system controls, a new dash-mounted control knob for the 4X4 system to replace the old pushbutton controls, brushed aluminum interior trim panels, new interior color options, available remote engine start, and a center dashboard-mounted analog clock.Also for 2008, a new 393-horsepower, 6.2 L V8 gasoline engine replaced the old 325-horsepower, 6.0 L V8 gasoline unit, with a new \"6L80-E\" six-speed automatic transmission replaced the old \"4L65-E\" four-speed automatic.(5) <image>Hummer H2. the time.The H2 was built by AM General under contract with General Motors at AM General Military Assembly Plant in Mishawaka, Indiana. 2002–2003 Vortec 6000 6.0 LQ4 (366 in³) V8- 316 (hp) at 5200 rpm- 0-60 times: 10.9 sec 2004–2007 Vortec 6000 6.0 LQ4 (366 in³) V8- 325 (hp) at 5200 rpm- 0-60 times: 10.7 sec 2008–2009 Vortec 6200 6.2 L92 V8- 393 (hp) at 5700 rpm- 0-60 times: 9.1 sec## Features.## Standard.Standard features include air conditioning with tri-zone climate controls, tilt leather-wrapped steering wheel with radio controls, cruise control, leather upholstery, heated front and rear seats, 8-way power front.\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/00/oven_00008190.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q618/Q618590.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q618/Q618590.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q618/Q618590.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q618/Q618590.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q618/Q618590.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 1,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q618/Q618590.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q618/Q618590.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q618/Q618590.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q618/Q618590.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q618/Q618590.jpg"
],
"candidate_texts": [
"Hummer H2. Air compressor with road assistance kit, and 20\" chrome wheels (slightly different from stock wheels).## 2008 updates.For 2008, the Hummer H2 and H2 SUT received an update. While largely unchanged on the exterior, the H2 and H2 SUT interiors were redesigned. This included new instrument cluster with improved gauges and Driver Information Center (DIC), a new leather-wrapped steering wheel, three new radios with a Bose premium audio system and auxiliary audio inputs (including a new touchscreen GPS navigation radio with DVD audio and video playback while in park, and XM Nav-Traffic capabilities), a new rear seat DVD entertainment system, Bluetooth.",
"Hummer H2. The Hummer H2 is a large SUV that was marketed by Hummer and built in the AM General facility under contract from General Motors from 2002 to 2009. It is based on a modified GMT820 GM three-quarter-ton pickup truck in the front and a half-ton 1500 frame in the rear. A four-door pickup truck version with a midgate that opens the vehicle's interior to the external cargo bed was introduced for 2005 as the H2 SUT (sport utility truck).## Background.In 2001 and 2002, GM allowed journalists to examine early versions of the Hummer H2, which were still under development at.",
"Hummer H2. seats, dual memory system, BOSE premium sound system, single-CD/cassette player and later in 2004, a 6-disc CD changer, then in 2008,a single CD-player with MP3 capability, an auxiliary input jack and DVD player, outside-temperature indicator, compass, rear radio controls, independent front torsion bar suspension, rear 5-link coil spring suspension specially created for the H2, oversized tires with HUMMER wheels, universal garage door opener and remote engine start (2008–2009).## Optional.Options for the H2 include adjustable rear suspension [which is included within the Adventure Package], a wide power sunroof, rearview camera, DVD entertainment system, navigation system, ladder, custom grilles, side step bars,.",
"Hummer H2. hands-free calling capabilities, enhanced voice activation, new OnStar hardware with buttons moved from the rearview mirror to the overhead console, a lower dashboard-mounted radio, new dual-zone climate controls, new rear seat audio system controls, a new dash-mounted control knob for the 4X4 system to replace the old pushbutton controls, brushed aluminum interior trim panels, new interior color options, available remote engine start, and a center dashboard-mounted analog clock.Also for 2008, a new 393-horsepower, 6.2 L V8 gasoline engine replaced the old 325-horsepower, 6.0 L V8 gasoline unit, with a new \"6L80-E\" six-speed automatic transmission replaced the old \"4L65-E\" four-speed automatic.",
"Hummer H2. the time.The H2 was built by AM General under contract with General Motors at AM General Military Assembly Plant in Mishawaka, Indiana. 2002–2003 Vortec 6000 6.0 LQ4 (366 in³) V8- 316 (hp) at 5200 rpm- 0-60 times: 10.9 sec 2004–2007 Vortec 6000 6.0 LQ4 (366 in³) V8- 325 (hp) at 5200 rpm- 0-60 times: 10.7 sec 2008–2009 Vortec 6200 6.2 L92 V8- 393 (hp) at 5700 rpm- 0-60 times: 9.1 sec## Features.## Standard.Standard features include air conditioning with tri-zone climate controls, tilt leather-wrapped steering wheel with radio controls, cruise control, leather upholstery, heated front and rear seats, 8-way power front."
],
"ground_truth_position": 1,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/00/oven_00008190.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q618/Q618590.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q618/Q618590.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q618/Q618590.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q618/Q618590.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q618/Q618590.jpg"
],
"num_candidates": 5,
"qid": "6:208493",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/00/oven_00008190.jpg",
"query_text": "What is the brand of this vehicle?"
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:Milwaukee Bucks forward Jabari Parker is one of the young stars of the team.Candidates:(1) <image>(2) <image>(3) <image>(4) <image>(5) <image>\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/washington_post/images/0490/719.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0704/791.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0622/295.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0199/941.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0020/811.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 5,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/washington_post/images/0490/719.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0704/791.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0622/295.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0199/941.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0020/811.jpg"
],
"candidate_texts": [],
"ground_truth_position": 5,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/washington_post/images/0490/719.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0704/791.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0622/295.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0199/941.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0020/811.jpg"
],
"num_candidates": 5,
"qid": "0:161978",
"query_image": null,
"query_text": "Milwaukee Bucks forward Jabari Parker is one of the young stars of the team."
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Shows another brown glass bottle with a person on the label.Candidates:(1) <image>(2) <image>(3) <image>(4) <image>(5) <image>\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/17/train-13189-0-img0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/90/train-6256-3-img0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/18/train-2304-3-img0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/88/train-6752-2-img1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/59/train-5908-1-img0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/63/train-9940-0-img0.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 4,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/90/train-6256-3-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/18/train-2304-3-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/88/train-6752-2-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/59/train-5908-1-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/63/train-9940-0-img0.jpg"
],
"candidate_texts": [],
"ground_truth_position": 4,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/17/train-13189-0-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/90/train-6256-3-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/18/train-2304-3-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/88/train-6752-2-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/59/train-5908-1-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/63/train-9940-0-img0.jpg"
],
"num_candidates": 5,
"qid": "8:20686",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/17/train-13189-0-img0.jpg",
"query_text": "Shows another brown glass bottle with a person on the label."
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Candidates:(1) A stop sign at a corner in the woods.(2) The stop sign is located at the intersection of Ajax and Ajax.(3) A red stop sign sitting under a couple of street signs.(4) A stop sign is positioned under two street signs with the name Ajax.(5) Street and stop signs direct traffic in the proper direction.\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000454232.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 1,
"style": "rule"
}
|
{
"candidate_images": [],
"candidate_texts": [
"A stop sign at a corner in the woods.",
"The stop sign is located at the intersection of Ajax and Ajax.",
"A red stop sign sitting under a couple of street signs.",
"A stop sign is positioned under two street signs with the name Ajax.",
"Street and stop signs direct traffic in the proper direction."
],
"ground_truth_position": 1,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000454232.jpg"
],
"num_candidates": 5,
"qid": "9:155359",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000454232.jpg",
"query_text": ""
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>What animal is presented in the image?Candidates:(1) Carphophis amoenus. Carphophis amoenus, commonly known as the worm snake, is a species of nonvenomous colubrid snake endemic to the eastern United States. \"C. amoenus\" can be found east of the Mississippi, from southwest Massachusetts south to southern Alabama west to Louisiana and then north to Illinois. This species of snake protects a large range, and normally prefers a moist habitat in the rocky woodlands, under rotten wood of logs and stumps. Though this snake is quite abundant over its range, it is rarely seen because of its dormant lifestyle and where it usually resides. This snake is most common on the.(2) Carphophis amoenus. Carphophis amoenus, commonly known as the worm snake, is a species of nonvenomous colubrid snake endemic to the eastern United States. \"C. amoenus\" can be found east of the Mississippi, from southwest Massachusetts south to southern Alabama west to Louisiana and then north to Illinois. This species of snake protects a large range, and normally prefers a moist habitat in the rocky woodlands, under rotten wood of logs and stumps. Though this snake is quite abundant over its range, it is rarely seen because of its dormant lifestyle and where it usually resides. This snake is most common on the.(3) Carphophis amoenus. edges or in the ecotonal areas of open to thick woodlands, and the borders of wetlands. It may also be found in the grasslands next to woodlands. The best chance to spot it is after heavy rains, when its small size and distinct color make it easy to spot. This species prefers moist soil inhabited by earthworms, which are its main prey, so the soil needs to be sufficiently moist. The snake's skin naturally evaporates water; so the soil needs to be moist enough to offset this. \"C. amoenus\" is mostly found under rocks and in sufficient leaf litter during.(4) Carphophis amoenus. short in comparison with its body and ends in an abrupt, spine-like scale. Females are longer than males, but have shorter tails. The head is small, conical and no wider than the neck. Other small, unpatterned brownish snakes, such as earth snakes (genus \"Virginia\") and red-bellied snakes (\"Storeria occipitomaculata\"), both have keeled body scales, but lack the spine-tipped tail. Other ways to distinguish between \"C. amoenus\" and other species is the body scales usually occur in 13 rows and are smooth and pitless, and the anal plate is split. Another snake commonly confused with \"C. amoenus\" is the western worm.(5) Carphophis amoenus. the extreme daytime heat.## Description.The worm snake is a small snake. Adults are 19 – 28 (cm) in total length, record 34 (cm) . It is brown dorsally, and bright pink ventrally, with the belly color including one or two dorsal scale rows. The dorsal scales are smooth, in 13 rows. It has five upper labials and one postocular. \"C. amoenus\" coloration is unpatterned and can be tan to dark brown in color; it has a pointed head, and small, black eyes. It has pinkish ventral pigmentation which extends dorsally onto dorsal scale rows one to two. The tail is.\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/00/oven_00952262.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 1,
"style": "rule"
}
|
{
"candidate_images": [],
"candidate_texts": [
"Carphophis amoenus. Carphophis amoenus, commonly known as the worm snake, is a species of nonvenomous colubrid snake endemic to the eastern United States. \"C. amoenus\" can be found east of the Mississippi, from southwest Massachusetts south to southern Alabama west to Louisiana and then north to Illinois. This species of snake protects a large range, and normally prefers a moist habitat in the rocky woodlands, under rotten wood of logs and stumps. Though this snake is quite abundant over its range, it is rarely seen because of its dormant lifestyle and where it usually resides. This snake is most common on the.",
"Carphophis amoenus. Carphophis amoenus, commonly known as the worm snake, is a species of nonvenomous colubrid snake endemic to the eastern United States. \"C. amoenus\" can be found east of the Mississippi, from southwest Massachusetts south to southern Alabama west to Louisiana and then north to Illinois. This species of snake protects a large range, and normally prefers a moist habitat in the rocky woodlands, under rotten wood of logs and stumps. Though this snake is quite abundant over its range, it is rarely seen because of its dormant lifestyle and where it usually resides. This snake is most common on the.",
"Carphophis amoenus. edges or in the ecotonal areas of open to thick woodlands, and the borders of wetlands. It may also be found in the grasslands next to woodlands. The best chance to spot it is after heavy rains, when its small size and distinct color make it easy to spot. This species prefers moist soil inhabited by earthworms, which are its main prey, so the soil needs to be sufficiently moist. The snake's skin naturally evaporates water; so the soil needs to be moist enough to offset this. \"C. amoenus\" is mostly found under rocks and in sufficient leaf litter during.",
"Carphophis amoenus. short in comparison with its body and ends in an abrupt, spine-like scale. Females are longer than males, but have shorter tails. The head is small, conical and no wider than the neck. Other small, unpatterned brownish snakes, such as earth snakes (genus \"Virginia\") and red-bellied snakes (\"Storeria occipitomaculata\"), both have keeled body scales, but lack the spine-tipped tail. Other ways to distinguish between \"C. amoenus\" and other species is the body scales usually occur in 13 rows and are smooth and pitless, and the anal plate is split. Another snake commonly confused with \"C. amoenus\" is the western worm.",
"Carphophis amoenus. the extreme daytime heat.## Description.The worm snake is a small snake. Adults are 19 – 28 (cm) in total length, record 34 (cm) . It is brown dorsally, and bright pink ventrally, with the belly color including one or two dorsal scale rows. The dorsal scales are smooth, in 13 rows. It has five upper labials and one postocular. \"C. amoenus\" coloration is unpatterned and can be tan to dark brown in color; it has a pointed head, and small, black eyes. It has pinkish ventral pigmentation which extends dorsally onto dorsal scale rows one to two. The tail is."
],
"ground_truth_position": 1,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/00/oven_00952262.jpg"
],
"num_candidates": 5,
"qid": "5:149488",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/00/oven_00952262.jpg",
"query_text": "What animal is presented in the image?"
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Add a couch to make the dog confortable.Candidates:(1) <image>(2) <image>(3) <image>(4) <image>(5) <image>\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/65/train-753-2-img0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/16/train-3366-0-img0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/43/train-6432-2-img0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/74/train-7480-1-img1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/9/train-11574-2-img0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/70/train-8480-0-img0.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 4,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/16/train-3366-0-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/43/train-6432-2-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/74/train-7480-1-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/9/train-11574-2-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/70/train-8480-0-img0.jpg"
],
"candidate_texts": [],
"ground_truth_position": 4,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/65/train-753-2-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/16/train-3366-0-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/43/train-6432-2-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/74/train-7480-1-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/9/train-11574-2-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/70/train-8480-0-img0.jpg"
],
"num_candidates": 5,
"qid": "8:22636",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/65/train-753-2-img0.jpg",
"query_text": "Add a couch to make the dog confortable."
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:Where are the tallest bookshelves in the Yale Law School Library Reading Room?Candidates:(1) <image>Cross Campus Highsmith View of Sterling Memorial Library and Cross Campus, Yale University, New Haven, CT.(2) <image>Reading Room, Langdell Hall, Harvard University, Cambridge MA Reading Room, Langdell Hall, Harvard University, Cambridge Massachusetts.(3) <image>Yale Law School Library Reading Room (L3) Yale Law School Library Reading Room (L3).(4) <image>Yale Law School Courtyard.(5) <image>Library of Congress Main Reading Room 2 As viewed from the tour room.\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30318117.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30299832.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30300017.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30375878.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30102439.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 3,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30318117.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30299832.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30300017.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30375878.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30102439.jpg"
],
"candidate_texts": [
"Cross Campus Highsmith View of Sterling Memorial Library and Cross Campus, Yale University, New Haven, CT.",
"Reading Room, Langdell Hall, Harvard University, Cambridge MA Reading Room, Langdell Hall, Harvard University, Cambridge Massachusetts.",
"Yale Law School Library Reading Room (L3) Yale Law School Library Reading Room (L3).",
"Yale Law School Courtyard.",
"Library of Congress Main Reading Room 2 As viewed from the tour room."
],
"ground_truth_position": 3,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30318117.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30299832.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30300017.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30375878.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30102439.jpg"
],
"num_candidates": 5,
"qid": "2:13236",
"query_image": null,
"query_text": "Where are the tallest bookshelves in the Yale Law School Library Reading Room?"
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>What is the closest parent taxonomy of this animal?Candidates:(1) Common garter snake. \"T. sirtalis\" that do not live in areas that contain \"T. granulosa\" contain the lowest levels of tetrodotoxin resistance, while those that do live in the same area have the highest levels of tetrodotoxin resistance. In populations where tetrodotoxin is absent in \"T. granulosa\", resistance in \"T. sirtalis\" is selected against because the mutation causes lower average population fitness. This helps maintain polymorphism within garter snake populations.## Subspecies.Current scientific classification recognizes 13 subspecies (ordered by date):A trinomial authority in parentheses indicates that the subspecies was originally described in a genus other than \"Thamnophis\".## Etymology.The subspecific name \"fitchi\" is in honor.(2) Common garter snake. The common garter snake (\"Thamnophis sirtalis\") is a species of natricine snake, which is indigenous to North America and found widely across the continent. Most common garter snakes have a pattern of yellow stripes on a black, brown or green background, and their average total length (including tail) is about 55 (cm), with a maximum total length of about 137 (cm). The average body mass is 150 (g). Common garter snakes are the state reptile of Massachusetts.## Description.Common garter snakes are thin snakes. Few grow over about 4 (ft) long, and most stay smaller. Most have longitudinal stripes in many.(3) Common garter snake. garter snake uses toxicity for both offense and defense. On the offensive side, the snake's venom can be toxic to some of its smaller prey, such as mice and other rodents. On the defensive side, the snake uses its resistance to toxicity to provide an important antipredator capability. A study on the evolutionary development of resistance of tetrodotoxin compared two populations of \"Thamnophis\" and then tested inside a population of \"T. sirtalis\". Those that were exposed to and lived in the same environment as the newts (\"Taricha granulosa\") or rough-skinned newt, that produce tetrodotoxin when eaten were more immune to.(4) Common garter snake. be toxic to amphibians and other small animals. Garter snakes have a mild venom in their saliva. For humans, a bite is not dangerous, though it may cause slight itching, burning, and/or swelling. Common garter snakes may also secrete a foul-smelling fluid from postanal glands when handled or harmed.Common garter snakes are resistant to naturally found poisons such as that of the American toad and rough-skinned newt, the latter of which can kill a human if ingested. They have the ability to absorb the toxin from the newts into their bodies, making them poisonous, which can deter potential predators.The common.(5) Common garter snake. American bullfrogs, common snapping turtles, larger snakes, hawks, raccoons, foxes, wild turkeys, and domestic cats and dogs.## Conservation.Water contamination, urban expansion, and residential and industrial development are all threats to the common garter snake. The San Francisco garter snake (\"T. s. tetrataenia\"), which is extremely scarce and occurs only in the vicinity of ponds and reservoirs in San Mateo County, California, has been listed as an endangered species by the U.S. Fish and Wildlife Service since 1967.## Antipredatory displays.Garter snakes exhibit many different behaviors to ward off predators. Garter snakes exhibit a greater variety of body postures than other snakes.\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/00/oven_00631994.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 1,
"style": "rule"
}
|
{
"candidate_images": [],
"candidate_texts": [
"Common garter snake. \"T. sirtalis\" that do not live in areas that contain \"T. granulosa\" contain the lowest levels of tetrodotoxin resistance, while those that do live in the same area have the highest levels of tetrodotoxin resistance. In populations where tetrodotoxin is absent in \"T. granulosa\", resistance in \"T. sirtalis\" is selected against because the mutation causes lower average population fitness. This helps maintain polymorphism within garter snake populations.## Subspecies.Current scientific classification recognizes 13 subspecies (ordered by date):A trinomial authority in parentheses indicates that the subspecies was originally described in a genus other than \"Thamnophis\".## Etymology.The subspecific name \"fitchi\" is in honor.",
"Common garter snake. The common garter snake (\"Thamnophis sirtalis\") is a species of natricine snake, which is indigenous to North America and found widely across the continent. Most common garter snakes have a pattern of yellow stripes on a black, brown or green background, and their average total length (including tail) is about 55 (cm), with a maximum total length of about 137 (cm). The average body mass is 150 (g). Common garter snakes are the state reptile of Massachusetts.## Description.Common garter snakes are thin snakes. Few grow over about 4 (ft) long, and most stay smaller. Most have longitudinal stripes in many.",
"Common garter snake. garter snake uses toxicity for both offense and defense. On the offensive side, the snake's venom can be toxic to some of its smaller prey, such as mice and other rodents. On the defensive side, the snake uses its resistance to toxicity to provide an important antipredator capability. A study on the evolutionary development of resistance of tetrodotoxin compared two populations of \"Thamnophis\" and then tested inside a population of \"T. sirtalis\". Those that were exposed to and lived in the same environment as the newts (\"Taricha granulosa\") or rough-skinned newt, that produce tetrodotoxin when eaten were more immune to.",
"Common garter snake. be toxic to amphibians and other small animals. Garter snakes have a mild venom in their saliva. For humans, a bite is not dangerous, though it may cause slight itching, burning, and/or swelling. Common garter snakes may also secrete a foul-smelling fluid from postanal glands when handled or harmed.Common garter snakes are resistant to naturally found poisons such as that of the American toad and rough-skinned newt, the latter of which can kill a human if ingested. They have the ability to absorb the toxin from the newts into their bodies, making them poisonous, which can deter potential predators.The common.",
"Common garter snake. American bullfrogs, common snapping turtles, larger snakes, hawks, raccoons, foxes, wild turkeys, and domestic cats and dogs.## Conservation.Water contamination, urban expansion, and residential and industrial development are all threats to the common garter snake. The San Francisco garter snake (\"T. s. tetrataenia\"), which is extremely scarce and occurs only in the vicinity of ponds and reservoirs in San Mateo County, California, has been listed as an endangered species by the U.S. Fish and Wildlife Service since 1967.## Antipredatory displays.Garter snakes exhibit many different behaviors to ward off predators. Garter snakes exhibit a greater variety of body postures than other snakes."
],
"ground_truth_position": 1,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/00/oven_00631994.jpg"
],
"num_candidates": 5,
"qid": "6:152426",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/00/oven_00631994.jpg",
"query_text": "What is the closest parent taxonomy of this animal?"
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Candidates:(1) <image>(2) <image>(3) <image>(4) <image>(5) <image>\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/ref/002/171.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/002/171_0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/001/222_1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/041/623_1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/095/135_1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/021/328_1.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 1,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/002/171_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/001/222_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/041/623_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/095/135_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/021/328_1.jpg"
],
"candidate_texts": [],
"ground_truth_position": 1,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/ref/002/171.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/002/171_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/001/222_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/041/623_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/095/135_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/021/328_1.jpg"
],
"num_candidates": 5,
"qid": "4:319",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/ref/002/171.jpg",
"query_text": ""
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:Multicolor textured straight leg joggers side stripes.Candidates:(1) <image>(2) <image>(3) <image>(4) <image>(5) <image>\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/pants/straight-leg_pants/91179865/91179865_1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/pants/full_length_pants/91111897/91111897_3.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/pants/straight-leg_pants/91179865/91179865_0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/pants/straight-leg_pants/91179865/91179865_3.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/pants/straight-leg_pants/91179865/91179865_2.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 1,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/pants/straight-leg_pants/91179865/91179865_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/pants/full_length_pants/91111897/91111897_3.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/pants/straight-leg_pants/91179865/91179865_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/pants/straight-leg_pants/91179865/91179865_3.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/pants/straight-leg_pants/91179865/91179865_2.jpg"
],
"candidate_texts": [],
"ground_truth_position": 1,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/pants/straight-leg_pants/91179865/91179865_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/pants/full_length_pants/91111897/91111897_3.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/pants/straight-leg_pants/91179865/91179865_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/pants/straight-leg_pants/91179865/91179865_3.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/pants/straight-leg_pants/91179865/91179865_2.jpg"
],
"num_candidates": 5,
"qid": "1:24924",
"query_image": null,
"query_text": "Multicolor textured straight leg joggers side stripes."
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Candidates:(1) <image>(2) <image>(3) <image>(4) <image>(5) <image>\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/ref/058/883.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/079/895_0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/010/385_0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/014/237_0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/058/883_0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/084/805_1.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 4,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/079/895_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/010/385_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/014/237_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/058/883_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/084/805_1.jpg"
],
"candidate_texts": [],
"ground_truth_position": 4,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/ref/058/883.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/079/895_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/010/385_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/014/237_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/058/883_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/084/805_1.jpg"
],
"num_candidates": 5,
"qid": "4:9382",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/ref/058/883.jpg",
"query_text": ""
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Candidates:(1) Black ruffled side top.(2) Multicolor puff sleeve top.(3) Black frilled flowing blouse.(4) Black ruffle sleeve cotton t-shirt.(5) White ruffle collar blouse.\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/tops/short_sleeve_tops/91269411/91269411_0.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 1,
"style": "rule"
}
|
{
"candidate_images": [],
"candidate_texts": [
"Black ruffled side top.",
"Multicolor puff sleeve top.",
"Black frilled flowing blouse.",
"Black ruffle sleeve cotton t-shirt.",
"White ruffle collar blouse."
],
"ground_truth_position": 1,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/tops/short_sleeve_tops/91269411/91269411_0.jpg"
],
"num_candidates": 5,
"qid": "1:27532",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/tops/short_sleeve_tops/91269411/91269411_0.jpg",
"query_text": ""
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:The Greek letter of X that is romanized as ch has a value of what number in Greek numerals?Candidates:(1) Chi is romanized as ⟨ch⟩ in most systematic transliteration conventions, but sometimes ⟨kh⟩ is used. In addition, in Modern Greek, it is often also romanized as ⟨h⟩ or ⟨x⟩ in informal practice. In the system of Greek numerals, it has a value of 600.(2) Rho / ˈroʊ / (uppercase Ρ, lowercase ρ or ϱ; Greek: ῥῶ) is the 17th letter of the Greek alphabet. In the system of Greek numerals it has a value of 100. It is derived from Phoenician letter res . Its uppercase form uses the same glyph, Ρ, as the distinct Latin letter P; the two letters have different Unicode encodings .(3) A number containing several decimal digits is built by appending the Roman numeral equivalent for each, from highest to lowest, as in the following examples:.(4) It is the origin of Greek Ϝ (digamma) and Υ (upsilon), Cyrillic У, Latin F and V, and the derived Latin- or Roman-alphabet letters U, W, and Y . The letter likely originated with an Egyptian hieroglyph which represented the word mace (transliterated as ḥ (dj)):.(5) Chi /ˈkaɪ/ (About this soundlisten)[1] (uppercase Χ, lowercase χ; Greek: χῖ) is the 22nd letter of the Greek alphabet. In the system of Greek numerals, it has a value of 600.\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[] |
multimodal_reranking
|
{
"ground_truth": 1,
"style": "rule"
}
|
{
"candidate_images": [],
"candidate_texts": [
"Chi is romanized as ⟨ch⟩ in most systematic transliteration conventions, but sometimes ⟨kh⟩ is used. In addition, in Modern Greek, it is often also romanized as ⟨h⟩ or ⟨x⟩ in informal practice. In the system of Greek numerals, it has a value of 600.",
"Rho / ˈroʊ / (uppercase Ρ, lowercase ρ or ϱ; Greek: ῥῶ) is the 17th letter of the Greek alphabet. In the system of Greek numerals it has a value of 100. It is derived from Phoenician letter res . Its uppercase form uses the same glyph, Ρ, as the distinct Latin letter P; the two letters have different Unicode encodings .",
"A number containing several decimal digits is built by appending the Roman numeral equivalent for each, from highest to lowest, as in the following examples:.",
"It is the origin of Greek Ϝ (digamma) and Υ (upsilon), Cyrillic У, Latin F and V, and the derived Latin- or Roman-alphabet letters U, W, and Y . The letter likely originated with an Egyptian hieroglyph which represented the word mace (transliterated as ḥ (dj)):.",
"Chi /ˈkaɪ/ (About this soundlisten)[1] (uppercase Χ, lowercase χ; Greek: χῖ) is the 22nd letter of the Greek alphabet. In the system of Greek numerals, it has a value of 600."
],
"ground_truth_position": 1,
"images": [],
"num_candidates": 5,
"qid": "2:26060",
"query_image": null,
"query_text": "The Greek letter of X that is romanized as ch has a value of what number in Greek numerals?"
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:A looted Debenhams store in Clapham Junction London on 8 August 2011.Candidates:(1) <image>4,558 claims have been made under the Riot Act in relation to last August's disturbances.(2) <image>Corrections & Clarifications: An earlier version of this story incorrectly stated that the National Guard had been deployed.(3) <image>27,310 Selfridges Photos and Premium High Res Pictures.(4) <image>Nine shootings were reported in a 24-hour period in Baltimore on Thursday, including at least two that were fatal.(5) <image>About 600 people were evacuated after a fire broke out at a market in Camden, north London.\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00187753_40000_03.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00139917_40000_04.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00189420_27247_00.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00774186_40000_03.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00982379_40000_02.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 1,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00187753_40000_03.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00139917_40000_04.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00189420_27247_00.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00774186_40000_03.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00982379_40000_02.jpg"
],
"candidate_texts": [
"4,558 claims have been made under the Riot Act in relation to last August's disturbances.",
"Corrections & Clarifications: An earlier version of this story incorrectly stated that the National Guard had been deployed.",
"27,310 Selfridges Photos and Premium High Res Pictures.",
"Nine shootings were reported in a 24-hour period in Baltimore on Thursday, including at least two that were fatal.",
"About 600 people were evacuated after a fire broke out at a market in Camden, north London."
],
"ground_truth_position": 1,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00187753_40000_03.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00139917_40000_04.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00189420_27247_00.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00774186_40000_03.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00982379_40000_02.jpg"
],
"num_candidates": 5,
"qid": "3:21110",
"query_image": null,
"query_text": "A looted Debenhams store in Clapham Junction London on 8 August 2011."
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:Black faux suede trim quilted jacket.Candidates:(1) <image>(2) <image>(3) <image>(4) <image>(5) <image>\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/jackets/casual_jackets/90551310/90551310_0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/jackets/leather_jackets/69378425/69378425_3.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/jackets/waistcoats_and_gilets/90797606/90797606_2.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/jackets/leather_jackets/90386635/90386635_0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/jackets/casual_jackets/91127672/91127672_2.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 1,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/jackets/casual_jackets/90551310/90551310_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/jackets/leather_jackets/69378425/69378425_3.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/jackets/waistcoats_and_gilets/90797606/90797606_2.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/jackets/leather_jackets/90386635/90386635_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/jackets/casual_jackets/91127672/91127672_2.jpg"
],
"candidate_texts": [],
"ground_truth_position": 1,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/jackets/casual_jackets/90551310/90551310_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/jackets/leather_jackets/69378425/69378425_3.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/jackets/waistcoats_and_gilets/90797606/90797606_2.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/jackets/leather_jackets/90386635/90386635_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/jackets/casual_jackets/91127672/91127672_2.jpg"
],
"num_candidates": 5,
"qid": "1:11951",
"query_image": null,
"query_text": "Black faux suede trim quilted jacket."
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Candidates:(1) Red moroccan tassel top.(2) Embroidered bell sleeve blouse.(3) Black embroidered gauze shift dress.(4) Black tie sleeve top.(5) Black eve geo burnout top.\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/dresses/mini_and_short_dresses/55639630/55639630_3.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 3,
"style": "rule"
}
|
{
"candidate_images": [],
"candidate_texts": [
"Red moroccan tassel top.",
"Embroidered bell sleeve blouse.",
"Black embroidered gauze shift dress.",
"Black tie sleeve top.",
"Black eve geo burnout top."
],
"ground_truth_position": 3,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/dresses/mini_and_short_dresses/55639630/55639630_3.jpg"
],
"num_candidates": 5,
"qid": "1:9727",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/dresses/mini_and_short_dresses/55639630/55639630_3.jpg",
"query_text": ""
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Show one large dessert with chocolate.Candidates:(1) <image>(2) <image>(3) <image>(4) <image>(5) <image>\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/41/train-9042-0-img1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/79/train-5143-1-img0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/24/train-8646-2-img0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/5/train-6637-0-img1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/40/train-11043-1-img1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/47/train-11917-3-img0.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 4,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/79/train-5143-1-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/24/train-8646-2-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/5/train-6637-0-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/40/train-11043-1-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/47/train-11917-3-img0.jpg"
],
"candidate_texts": [],
"ground_truth_position": 4,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/41/train-9042-0-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/79/train-5143-1-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/24/train-8646-2-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/5/train-6637-0-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/40/train-11043-1-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/47/train-11917-3-img0.jpg"
],
"num_candidates": 5,
"qid": "8:5684",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/41/train-9042-0-img1.jpg",
"query_text": "Show one large dessert with chocolate."
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Where is this place?Candidates:(1) <image>Blue Mosque, Tabriz. The Blue Mosque () is a historic mosque in Tabriz, Iran. The mosque and some other public buildings were constructed in 1465 upon the order of Jahan Shah, the ruler of Kara Koyunlu.The mosque was severely damaged in an earthquake in 1780, leaving only the iwan (entrance hall). Reconstruction began in 1973 by Reza Memaran Benam under the supervision of Iranian Ministry of Culture. However, it is still incomplete.## History.The Blue mosque of Tabriz was built upon the order of Jahan Shah, the ruler of the Kara Koyunlu. Jahan Shah's wife, Khatun Jan Begum (died 1469), established the endowment (\"vaqf\").(2) <image>Blue Mosque, Tabriz. The Blue Mosque () is a historic mosque in Tabriz, Iran. The mosque and some other public buildings were constructed in 1465 upon the order of Jahan Shah, the ruler of Kara Koyunlu.The mosque was severely damaged in an earthquake in 1780, leaving only the iwan (entrance hall). Reconstruction began in 1973 by Reza Memaran Benam under the supervision of Iranian Ministry of Culture. However, it is still incomplete.## History.The Blue mosque of Tabriz was built upon the order of Jahan Shah, the ruler of the Kara Koyunlu. Jahan Shah's wife, Khatun Jan Begum (died 1469), established the endowment (\"vaqf\").(3) <image>Blue Mosque, Tabriz. capture and occupation of the Blue Mosque, several earthquakes did damage the building between the 16th and 18th centuries. It was especially severely damaged by the earthquake of 1780. However, in the 17th century, the Blue Mosque was already reportedly \"completely destroyed and abandoned\". In the 19th century, the local people of Tabriz looted the building's ruins. In the 20th century, during the Pahlavi era, the mosque was finally rebuilt. Reconstruction is performed by plans and supervision of Mohammad Reza Memaran Benam, a traditional architect from Tabriz, with authority of the Iranian Organization of Cultural Heritages.## Calligraphy.The diverse Kufic, and.(4) <image>Blue Mosque, Tabriz. and Wilber, p. 407; Aube, p. 248)\".Though the mausoleum was never completed, when the Safavids assumed control over Tabriz and made it their capital, the Blue Mosque itself served the new rulers as a mosque during the first half of the 16th century. In 1514, after the Safavids were defeated at the decisive Battle of Chaldiran, the Ottomans occupied and looted Tabriz, including the Blue Mosque. Aube notes that at least eight carpets were looted by the Turks and taken to Istanbul. Aube notes that even though it is not known whether the Turks attacked the structure itself during the.(5) <image>Blue Mosque, Tabriz. for the mosque's construction. However, just a few years later, Jahan Shah and his Kara Koyunlu were toppled by Uzun Hassan of the Ak Koyunlu, and Tabriz was taken. Jahan Shah's daughter, Saleha Khatun, oversaw the rest of the construction work by the new rulers. During the reign of Yaqub bin Uzun Hasan, \"the cupola of the mosque's mausoleum as well as its main parts were completed\". Sandra Aube adds: \"A few details from the mausoleum’s interior, such as alabaster pieces from the wall panels and the main prayer niche (\"meḥrāb\"), reveal that the mausoleum was never completely finished (Golombek.\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/00/oven_00104420.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q151/Q1519733.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q151/Q1519733.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q151/Q1519733.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q151/Q1519733.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q151/Q1519733.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 1,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q151/Q1519733.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q151/Q1519733.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q151/Q1519733.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q151/Q1519733.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q151/Q1519733.jpg"
],
"candidate_texts": [
"Blue Mosque, Tabriz. The Blue Mosque () is a historic mosque in Tabriz, Iran. The mosque and some other public buildings were constructed in 1465 upon the order of Jahan Shah, the ruler of Kara Koyunlu.The mosque was severely damaged in an earthquake in 1780, leaving only the iwan (entrance hall). Reconstruction began in 1973 by Reza Memaran Benam under the supervision of Iranian Ministry of Culture. However, it is still incomplete.## History.The Blue mosque of Tabriz was built upon the order of Jahan Shah, the ruler of the Kara Koyunlu. Jahan Shah's wife, Khatun Jan Begum (died 1469), established the endowment (\"vaqf\").",
"Blue Mosque, Tabriz. The Blue Mosque () is a historic mosque in Tabriz, Iran. The mosque and some other public buildings were constructed in 1465 upon the order of Jahan Shah, the ruler of Kara Koyunlu.The mosque was severely damaged in an earthquake in 1780, leaving only the iwan (entrance hall). Reconstruction began in 1973 by Reza Memaran Benam under the supervision of Iranian Ministry of Culture. However, it is still incomplete.## History.The Blue mosque of Tabriz was built upon the order of Jahan Shah, the ruler of the Kara Koyunlu. Jahan Shah's wife, Khatun Jan Begum (died 1469), established the endowment (\"vaqf\").",
"Blue Mosque, Tabriz. capture and occupation of the Blue Mosque, several earthquakes did damage the building between the 16th and 18th centuries. It was especially severely damaged by the earthquake of 1780. However, in the 17th century, the Blue Mosque was already reportedly \"completely destroyed and abandoned\". In the 19th century, the local people of Tabriz looted the building's ruins. In the 20th century, during the Pahlavi era, the mosque was finally rebuilt. Reconstruction is performed by plans and supervision of Mohammad Reza Memaran Benam, a traditional architect from Tabriz, with authority of the Iranian Organization of Cultural Heritages.## Calligraphy.The diverse Kufic, and.",
"Blue Mosque, Tabriz. and Wilber, p. 407; Aube, p. 248)\".Though the mausoleum was never completed, when the Safavids assumed control over Tabriz and made it their capital, the Blue Mosque itself served the new rulers as a mosque during the first half of the 16th century. In 1514, after the Safavids were defeated at the decisive Battle of Chaldiran, the Ottomans occupied and looted Tabriz, including the Blue Mosque. Aube notes that at least eight carpets were looted by the Turks and taken to Istanbul. Aube notes that even though it is not known whether the Turks attacked the structure itself during the.",
"Blue Mosque, Tabriz. for the mosque's construction. However, just a few years later, Jahan Shah and his Kara Koyunlu were toppled by Uzun Hassan of the Ak Koyunlu, and Tabriz was taken. Jahan Shah's daughter, Saleha Khatun, oversaw the rest of the construction work by the new rulers. During the reign of Yaqub bin Uzun Hasan, \"the cupola of the mosque's mausoleum as well as its main parts were completed\". Sandra Aube adds: \"A few details from the mausoleum’s interior, such as alabaster pieces from the wall panels and the main prayer niche (\"meḥrāb\"), reveal that the mausoleum was never completely finished (Golombek."
],
"ground_truth_position": 1,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/00/oven_00104420.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q151/Q1519733.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q151/Q1519733.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q151/Q1519733.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q151/Q1519733.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q151/Q1519733.jpg"
],
"num_candidates": 5,
"qid": "5:76654",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/00/oven_00104420.jpg",
"query_text": "Where is this place?"
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Put the bowl of mashed potatoes on a gray napkin.Candidates:(1) <image>(2) <image>(3) <image>(4) <image>(5) <image>\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/48/train-1645-3-img0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/62/train-1947-1-img1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/58/train-11397-0-img0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/20/train-1481-2-img1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/75/train-5806-3-img0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/50/train-9119-0-img1.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 3,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/62/train-1947-1-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/58/train-11397-0-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/20/train-1481-2-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/75/train-5806-3-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/50/train-9119-0-img1.jpg"
],
"candidate_texts": [],
"ground_truth_position": 3,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/48/train-1645-3-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/62/train-1947-1-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/58/train-11397-0-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/20/train-1481-2-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/75/train-5806-3-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/50/train-9119-0-img1.jpg"
],
"num_candidates": 5,
"qid": "8:10170",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/48/train-1645-3-img0.jpg",
"query_text": "Put the bowl of mashed potatoes on a gray napkin."
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Candidates:(1) About 100 people gathered at Westlake Park in Seattle as the grand jury decision was announced then took to the streets in protest.(2) A protester stands on the steps of City Hall Friday in March in San Francisco A Justice Department review released Wednesday from the San Francisco Police Department stops and searches African Americans at a disproportionately high rate and does not adequately investigate officers using force.(3) Epa04493669 Protesters march down the streets in Clayton Missouri USA 17 November 2014 People wait for the Grand Jury verdict over the police slaying of an unarmed black teenager in Ferguson Missouri Brown s death on 09 August set off weeks of sometimes violent demonstrations in Ferguson as protesters demanded that prosecutors charge police officer Darren Wilson who shot the 18yearold six times EPALARRY W SMITH.(4) Hundreds of people march to the headquarters of the Albuquerque Police Department on March 25 2013 protesting the police killing of James Boyd in the northeast foothills of Albuquerque.(5) Albuquerque police face off with protesters Sunday in downtown Albuquerque NM during a protest against recent police shootings.\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0203/448.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 4,
"style": "rule"
}
|
{
"candidate_images": [],
"candidate_texts": [
"About 100 people gathered at Westlake Park in Seattle as the grand jury decision was announced then took to the streets in protest.",
"A protester stands on the steps of City Hall Friday in March in San Francisco A Justice Department review released Wednesday from the San Francisco Police Department stops and searches African Americans at a disproportionately high rate and does not adequately investigate officers using force.",
"Epa04493669 Protesters march down the streets in Clayton Missouri USA 17 November 2014 People wait for the Grand Jury verdict over the police slaying of an unarmed black teenager in Ferguson Missouri Brown s death on 09 August set off weeks of sometimes violent demonstrations in Ferguson as protesters demanded that prosecutors charge police officer Darren Wilson who shot the 18yearold six times EPALARRY W SMITH.",
"Hundreds of people march to the headquarters of the Albuquerque Police Department on March 25 2013 protesting the police killing of James Boyd in the northeast foothills of Albuquerque.",
"Albuquerque police face off with protesters Sunday in downtown Albuquerque NM during a protest against recent police shootings."
],
"ground_truth_position": 4,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0203/448.jpg"
],
"num_candidates": 5,
"qid": "0:199424",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0203/448.jpg",
"query_text": ""
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:Donald Trump fields a question during the first Republican presidential debate hosted by Fox News and Facebook on Thursday in Cleveland.Candidates:(1) <image>2016 Presidential Debates: Latest News, Top Stories & Analysis- POLITICO.(2) <image>Why Donald Trump Will Still Dominate the Republican Debate.(3) <image>A tough night for Trump and political correctness: 5 takeaways from the Republican debate.(4) <image>Transcript of Republican Presidential Debate.(5) <image>The FixDon t understand the Trump phenomenon? Just watch Megyn Kelly s voter panel.\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00161563_23763_00.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00155802_23043_00.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00097686_15741_00.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00132256_20092_00.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00100540_40000_05.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 5,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00161563_23763_00.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00155802_23043_00.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00097686_15741_00.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00132256_20092_00.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00100540_40000_05.jpg"
],
"candidate_texts": [
"2016 Presidential Debates: Latest News, Top Stories & Analysis- POLITICO.",
"Why Donald Trump Will Still Dominate the Republican Debate.",
"A tough night for Trump and political correctness: 5 takeaways from the Republican debate.",
"Transcript of Republican Presidential Debate.",
"The FixDon t understand the Trump phenomenon? Just watch Megyn Kelly s voter panel."
],
"ground_truth_position": 5,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00161563_23763_00.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00155802_23043_00.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00097686_15741_00.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00132256_20092_00.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00100540_40000_05.jpg"
],
"num_candidates": 5,
"qid": "3:6562",
"query_image": null,
"query_text": "Donald Trump fields a question during the first Republican presidential debate hosted by Fox News and Facebook on Thursday in Cleveland."
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:A man comforts his daughter as she is treated for food poisoning in the Hassan Sham U2 camp for displaced people, east of Mosul, Iraq, Tuesday.Candidates:(1) <image>Hundreds of Iraqis Sickened by Food Poisoning at Camp Near Mosul.(2) <image>In Pictures: Iraq's children seek safety - Humanitarian Crises.(3) <image>Most Yazidis have been rescued from a besieged mountain in northern Iraq.(4) <image>Why Are Iraqi Kurds Fleeing to Europe?(5) <image>Iraq has done 'absolutely nothing' for displaced Christians.\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00047009_10676_01.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00127889_19545_00.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00106127_16806_00.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00124305_19095_00.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00085824_14248_00.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 1,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00047009_10676_01.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00127889_19545_00.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00106127_16806_00.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00124305_19095_00.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00085824_14248_00.jpg"
],
"candidate_texts": [
"Hundreds of Iraqis Sickened by Food Poisoning at Camp Near Mosul.",
"In Pictures: Iraq's children seek safety - Humanitarian Crises.",
"Most Yazidis have been rescued from a besieged mountain in northern Iraq.",
"Why Are Iraqi Kurds Fleeing to Europe?",
"Iraq has done 'absolutely nothing' for displaced Christians."
],
"ground_truth_position": 1,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00047009_10676_01.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00127889_19545_00.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00106127_16806_00.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00124305_19095_00.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00085824_14248_00.jpg"
],
"num_candidates": 5,
"qid": "3:13000",
"query_image": null,
"query_text": "A man comforts his daughter as she is treated for food poisoning in the Hassan Sham U2 camp for displaced people, east of Mosul, Iraq, Tuesday."
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Candidates:(1) <image>(2) <image>(3) <image>(4) <image>(5) <image>\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/ref/010/190.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/057/189_1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/054/627_0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/027/939_0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/010/190_0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/053/639_0.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 4,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/057/189_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/054/627_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/027/939_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/010/190_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/053/639_0.jpg"
],
"candidate_texts": [],
"ground_truth_position": 4,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/ref/010/190.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/057/189_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/054/627_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/027/939_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/010/190_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/053/639_0.jpg"
],
"num_candidates": 5,
"qid": "4:1621",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/ref/010/190.jpg",
"query_text": ""
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:What artistic endeavor do David Kos Rolfe and Akos Kovacs both pursue?Candidates:(1) The nature of Nine Inch Nails frontman Trent Reznor 's contribution to 1000 Homo DJs' records has been debated. What is certain is that Reznor recorded the original vocals for \" Supernaut .\" This performance was not officially used because Reznor's label, TVT Records, refused to allow his appearance on the release.(2) Beauford would later recall that, \"It started out as a three-piece thing with Dave and Leroi...working on some of Dave's songs. He only had four songs at the time..(3) The native form of this personal name is Kovács Ákos . This article uses Western name order when mentioning individuals. Ákos Kovács (born in Budapest, 6 April 1968) is a Hungarian pop - rock singer-songwriter.(4) David Kos \"Music Man\" Rolfe (born November 23, 1968) is an American singer from New York City who has written and performed 5 English songs for the Pokémon anime. Rolfe is currently working at advertising agency BBDO and is still continues to make music as of 2017. David Rolfe was born in New York on November 23, 1968.(5) Vangelis (born March 29, 1943) is a Greek composer and performer. He gained wide popularity after composing soundtracks to Chariots of Fire (1981) and Blade Runner (1982).\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[] |
multimodal_reranking
|
{
"ground_truth": 3,
"style": "rule"
}
|
{
"candidate_images": [],
"candidate_texts": [
"The nature of Nine Inch Nails frontman Trent Reznor 's contribution to 1000 Homo DJs' records has been debated. What is certain is that Reznor recorded the original vocals for \" Supernaut .\" This performance was not officially used because Reznor's label, TVT Records, refused to allow his appearance on the release.",
"Beauford would later recall that, \"It started out as a three-piece thing with Dave and Leroi...working on some of Dave's songs. He only had four songs at the time..",
"The native form of this personal name is Kovács Ákos . This article uses Western name order when mentioning individuals. Ákos Kovács (born in Budapest, 6 April 1968) is a Hungarian pop - rock singer-songwriter.",
"David Kos \"Music Man\" Rolfe (born November 23, 1968) is an American singer from New York City who has written and performed 5 English songs for the Pokémon anime. Rolfe is currently working at advertising agency BBDO and is still continues to make music as of 2017. David Rolfe was born in New York on November 23, 1968.",
"Vangelis (born March 29, 1943) is a Greek composer and performer. He gained wide popularity after composing soundtracks to Chariots of Fire (1981) and Blade Runner (1982)."
],
"ground_truth_position": 3,
"images": [],
"num_candidates": 5,
"qid": "2:23071",
"query_image": null,
"query_text": "What artistic endeavor do David Kos Rolfe and Akos Kovacs both pursue?"
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Three transparent glass measuring containers of different sizes on a blue background instead of five transparent glass measuring containers on a gray background.Candidates:(1) <image>(2) <image>(3) <image>(4) <image>(5) <image>\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/83/train-11167-2-img0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/31/train-11737-1-img0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/41/train-9042-1-img0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/89/train-8052-2-img1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/48/train-3249-3-img1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/88/train-10838-0-img1.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 1,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/31/train-11737-1-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/41/train-9042-1-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/89/train-8052-2-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/48/train-3249-3-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/88/train-10838-0-img1.jpg"
],
"candidate_texts": [],
"ground_truth_position": 1,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/83/train-11167-2-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/31/train-11737-1-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/41/train-9042-1-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/89/train-8052-2-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/48/train-3249-3-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/88/train-10838-0-img1.jpg"
],
"num_candidates": 5,
"qid": "8:18822",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/83/train-11167-2-img0.jpg",
"query_text": "Three transparent glass measuring containers of different sizes on a blue background instead of five transparent glass measuring containers on a gray background."
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>What product does this animal produce?Candidates:(1) Dairy. superette. This usage is historical as such shops were a common place for the public to buy milk products.## History.Milk producing animals have been domesticated for thousands of years. Initially, they were part of the subsistence farming that nomads engaged in. As the community moved about the country, their animals accompanied them. Protecting and feeding the animals were a big part of the symbiotic relationship between the animals and the herders.In the more recent past, people in agricultural societies owned dairy animals that they milked for domestic and local (village) consumption, a typical example of a cottage industry. The animals.(2) Dairy. (usually by truck) to a \"dairy plant\", also referred to as a \"dairy\", where raw milk is further processed and prepared for commercial sale of dairy products. In New Zealand, farm areas for milk harvesting are also called \"milking parlours\", and are historically known as \"milking sheds\". As in the United States, sometimes milking sheds are referred to by their type, such as \"herring bone shed\" or \"pit parlour\". Parlour design has evolved from simple barns or sheds to large rotary structures in which the workflow (throughput of cows) is very efficiently handled. In some countries, especially those with small.(3) Dairy. numbers of animals being milked, the farm may perform the functions of a dairy plant, processing their own milk into salable dairy products, such as butter, cheese, or yogurt. This on-site processing is a traditional method of producing specialist milk products, common in Europe.In the United States a \"dairy\" can also be a place that processes, distributes and sells dairy products, or a room, building or establishment where milk is stored and processed into milk products, such as butter or cheese. In New Zealand English the singular use of the word \"dairy\" almost exclusively refers to a corner shop, or.(4) Dairy. processes it into a variety of dairy products. These establishments constitute the global dairy industry, part of the food industry.## Terminology.Terminology differs between countries. In the United States, for example, an entire dairy farm is commonly called a \"dairy\". The building or farm area where milk is harvested from the cow is often called a \"milking parlor\" or \"parlor\", except in the case of smaller dairies, where cows are often put on pasture, and usually milked in \"stanchion barns\". The farm area where milk is stored in bulk tanks is known as the farm's \"milk house\". Milk is then hauled.(5) Dairy. in the milk that was trapped in the milk duct being squirted out the end into a bucket that is supported between the knees (or rests on the ground) of the milker, who usually sits on a low stool.Traditionally the cow, or cows, would stand in the field or paddock while being milked. Young stock, heifers, would have to be trained to remain still to be milked. In many countries, the cows were tethered to a post and milked.## Structure of the industry.While most countries produce their own milk products, the structure of the dairy industry varies in different parts.\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/02/oven_02572911.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 1,
"style": "rule"
}
|
{
"candidate_images": [],
"candidate_texts": [
"Dairy. superette. This usage is historical as such shops were a common place for the public to buy milk products.## History.Milk producing animals have been domesticated for thousands of years. Initially, they were part of the subsistence farming that nomads engaged in. As the community moved about the country, their animals accompanied them. Protecting and feeding the animals were a big part of the symbiotic relationship between the animals and the herders.In the more recent past, people in agricultural societies owned dairy animals that they milked for domestic and local (village) consumption, a typical example of a cottage industry. The animals.",
"Dairy. (usually by truck) to a \"dairy plant\", also referred to as a \"dairy\", where raw milk is further processed and prepared for commercial sale of dairy products. In New Zealand, farm areas for milk harvesting are also called \"milking parlours\", and are historically known as \"milking sheds\". As in the United States, sometimes milking sheds are referred to by their type, such as \"herring bone shed\" or \"pit parlour\". Parlour design has evolved from simple barns or sheds to large rotary structures in which the workflow (throughput of cows) is very efficiently handled. In some countries, especially those with small.",
"Dairy. numbers of animals being milked, the farm may perform the functions of a dairy plant, processing their own milk into salable dairy products, such as butter, cheese, or yogurt. This on-site processing is a traditional method of producing specialist milk products, common in Europe.In the United States a \"dairy\" can also be a place that processes, distributes and sells dairy products, or a room, building or establishment where milk is stored and processed into milk products, such as butter or cheese. In New Zealand English the singular use of the word \"dairy\" almost exclusively refers to a corner shop, or.",
"Dairy. processes it into a variety of dairy products. These establishments constitute the global dairy industry, part of the food industry.## Terminology.Terminology differs between countries. In the United States, for example, an entire dairy farm is commonly called a \"dairy\". The building or farm area where milk is harvested from the cow is often called a \"milking parlor\" or \"parlor\", except in the case of smaller dairies, where cows are often put on pasture, and usually milked in \"stanchion barns\". The farm area where milk is stored in bulk tanks is known as the farm's \"milk house\". Milk is then hauled.",
"Dairy. in the milk that was trapped in the milk duct being squirted out the end into a bucket that is supported between the knees (or rests on the ground) of the milker, who usually sits on a low stool.Traditionally the cow, or cows, would stand in the field or paddock while being milked. Young stock, heifers, would have to be trained to remain still to be milked. In many countries, the cows were tethered to a post and milked.## Structure of the industry.While most countries produce their own milk products, the structure of the dairy industry varies in different parts."
],
"ground_truth_position": 1,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/02/oven_02572911.jpg"
],
"num_candidates": 5,
"qid": "6:206994",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/02/oven_02572911.jpg",
"query_text": "What product does this animal produce?"
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Candidates:(1) The A27 was completely closed in both directions after the crash.(2) The A708 in southern Scotland has reopened a little more than a week after a section collapsed.(3) Fire crews and police were called to the scene after the crash.(4) Aerial photographs of the crash show its severity.(5) The road bridge was closed to all traffic shortly before 1700.\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/bbc/images/0135/397.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 1,
"style": "rule"
}
|
{
"candidate_images": [],
"candidate_texts": [
"The A27 was completely closed in both directions after the crash.",
"The A708 in southern Scotland has reopened a little more than a week after a section collapsed.",
"Fire crews and police were called to the scene after the crash.",
"Aerial photographs of the crash show its severity.",
"The road bridge was closed to all traffic shortly before 1700."
],
"ground_truth_position": 1,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/bbc/images/0135/397.jpg"
],
"num_candidates": 5,
"qid": "0:137449",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/bbc/images/0135/397.jpg",
"query_text": ""
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:A train is sitting on the railroad track.Candidates:(1) <image>(2) <image>(3) <image>(4) <image>(5) <image>\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000254407.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000245392.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000149731.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000330432.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000033758.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 4,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000254407.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000245392.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000149731.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000330432.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000033758.jpg"
],
"candidate_texts": [],
"ground_truth_position": 4,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000254407.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000245392.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000149731.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000330432.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000033758.jpg"
],
"num_candidates": 5,
"qid": "9:57162",
"query_image": null,
"query_text": "A train is sitting on the railroad track."
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Candidates:(1) <image>(2) <image>(3) <image>(4) <image>(5) <image>\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/ref/042/685.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/042/685_0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/067/394_1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/007/462_1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/067/394_0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/010/910_0.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 1,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/042/685_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/067/394_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/007/462_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/067/394_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/010/910_0.jpg"
],
"candidate_texts": [],
"ground_truth_position": 1,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/ref/042/685.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/042/685_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/067/394_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/007/462_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/067/394_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/010/910_0.jpg"
],
"num_candidates": 5,
"qid": "4:6815",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/ref/042/685.jpg",
"query_text": ""
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:Metropolitan Transportation Authority workers at an L train stop on Bedford Avenue in Brooklyn on Tuesday.Candidates:(1) <image>The real reasons to fear Cuomo's L train plan.(2) <image>7 Ways to Fix the M.T.A. (Which Needs a $60 Billion Overhaul).(3) <image>Some Brooklyn Riders Are Left Behind in Revival of the Subway System.(4) <image>NYC Subway Delays: Power Outage Causes 'Extensive' Problems, Service Changes Tuesday.(5) <image>MTA announces comprehensive plan to tackle subway delays.\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00032420_10221_00.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00065174_07316_01.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00032417_10221_01.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00004157_05403_00.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00048013_10712_00.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 1,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00032420_10221_00.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00065174_07316_01.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00032417_10221_01.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00004157_05403_00.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00048013_10712_00.jpg"
],
"candidate_texts": [
"The real reasons to fear Cuomo's L train plan.",
"7 Ways to Fix the M.T.A. (Which Needs a $60 Billion Overhaul).",
"Some Brooklyn Riders Are Left Behind in Revival of the Subway System.",
"NYC Subway Delays: Power Outage Causes 'Extensive' Problems, Service Changes Tuesday.",
"MTA announces comprehensive plan to tackle subway delays."
],
"ground_truth_position": 1,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00032420_10221_00.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00065174_07316_01.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00032417_10221_01.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00004157_05403_00.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/edis_images/00048013_10712_00.jpg"
],
"num_candidates": 5,
"qid": "3:10185",
"query_image": null,
"query_text": "Metropolitan Transportation Authority workers at an L train stop on Bedford Avenue in Brooklyn on Tuesday."
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>More focused on the firt dogs pulling the car and all the others in the background.Candidates:(1) <image>(2) <image>(3) <image>(4) <image>(5) <image>\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/97/train-6252-1-img1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/83/train-1119-3-img0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/73/train-4923-1-img1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/48/train-10949-0-img1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/36/train-1046-3-img1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/73/train-9675-0-img1.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 1,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/83/train-1119-3-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/73/train-4923-1-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/48/train-10949-0-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/36/train-1046-3-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/73/train-9675-0-img1.jpg"
],
"candidate_texts": [],
"ground_truth_position": 1,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/97/train-6252-1-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/83/train-1119-3-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/73/train-4923-1-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/48/train-10949-0-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/36/train-1046-3-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/73/train-9675-0-img1.jpg"
],
"num_candidates": 5,
"qid": "8:20906",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/97/train-6252-1-img1.jpg",
"query_text": "More focused on the firt dogs pulling the car and all the others in the background."
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Is more colorful and whimsical and has no sleeves and is patterned blue.Candidates:(1) <image>(2) <image>(3) <image>(4) <image>(5) <image>\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashioniq_images/B00BQTQBVG.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashioniq_images/B0020MZ2EU.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashioniq_images/B008OENU74.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashioniq_images/B008SRLABK.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashioniq_images/B007IUTGL4.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashioniq_images/B000JCBB2U.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 5,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashioniq_images/B0020MZ2EU.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashioniq_images/B008OENU74.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashioniq_images/B008SRLABK.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashioniq_images/B007IUTGL4.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashioniq_images/B000JCBB2U.jpg"
],
"candidate_texts": [],
"ground_truth_position": 5,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashioniq_images/B00BQTQBVG.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashioniq_images/B0020MZ2EU.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashioniq_images/B008OENU74.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashioniq_images/B008SRLABK.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashioniq_images/B007IUTGL4.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashioniq_images/B000JCBB2U.jpg"
],
"num_candidates": 5,
"qid": "7:15405",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashioniq_images/B00BQTQBVG.jpg",
"query_text": "Is more colorful and whimsical and has no sleeves and is patterned blue."
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Candidates:(1) <image>(2) <image>(3) <image>(4) <image>(5) <image>\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/ref/036/989.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/095/494_0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/054/620_1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/036/989_1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/078/217_0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/024/630_0.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 3,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/095/494_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/054/620_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/036/989_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/078/217_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/024/630_0.jpg"
],
"candidate_texts": [],
"ground_truth_position": 3,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/ref/036/989.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/095/494_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/054/620_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/036/989_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/078/217_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/024/630_0.jpg"
],
"num_candidates": 5,
"qid": "4:5882",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/ref/036/989.jpg",
"query_text": ""
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>What is the manufacturer of this aircraft?Candidates:(1) <image>Dassault Falcon 900. for dropping rescue stores.Falcon 900BRevised production version from 1991. Powered by 21.13 kN (4,750 lbf) TFE731-5BR-1C engines.Falcon 900CReplacement for 900B. Introduced in 2000.Long range version, with 22.24 kN (5,000 lbf) engines. This variant features TFE731-60 engines, with a range of 8,340 km (4,501 nm; 5,180 miles). Avionics by (Honeywell Primus). It entered service in 1996.Falcon 900EX EASyFalcon 900DXShorter-range production type. TFE731-60 engines.Falcon 900LXCurrent production variant of EX fitted with blended winglets. Range of 4750 (nmi).VC-900AItalian military designation for the 900EX.VC-900BItalian military designation for the 900EASy.## Operators.## Civil operators.A wide range of private owners, businesses and small airlines operate Falcon 900s.(2) <image>Boeing. gas emissions was poised to grow, as air travel increases and ground vehicles use more alternative fuels like ethanol and biodiesel. Boeing estimates that biofuels could reduce flight-related greenhouse-gas emissions by 60 to 80%. The solution blends algae fuels with existing jet fuel.Boeing executives said the company was collaborating with Brazilian biofuels maker Tecbio, Aquaflow Bionomic of New Zealand, and other fuel developers around the world. As of 2007, Boeing had tested six fuels from these companies, and expected to test 20 fuels \"by the time we're done evaluating them\". Boeing also joined other aviation-related members in the Algal Biomass.(3) <image>Boeing. 25% Hydro-treated Renewable Jet fuel and 25% of a Fischer–Tropsch fuel with successful results.## Electric propulsion.For NASA's N+3 future airliner program, Boeing has determined that hybrid electric engine technology is by far the best choice for its subsonic design. Hybrid electric propulsion has the potential to shorten takeoff distance and reduce noise. Boeing created a team to study electric propulsion in future generation of subsonic commercial aircraft. SUGAR for Subsonic Ultra Green Aircraft Research includes, BR&T, Boeing Commercial Airplanes, General Electric, and Georgia Tech. There are five main concepts the team is reviewing. SUGAR Free and Refined SUGAR, are two.(4) <image>Boeing. The Boeing Company () is an American multinational corporation that designs, manufactures, and sells airplanes, rotorcraft, rockets, satellites, telecommunications equipment, and missiles worldwide. The company also provides leasing and product support services. Boeing is among the largest global aerospace manufacturers; it is the third-largest defense contractor in the world based on 2020 revenue, and is the largest exporter in the United States by dollar value. Boeing stock is included in the Dow Jones Industrial Average. Boeing is incorporated in Delaware.Boeing was founded by William Boeing in Seattle, Washington, on July 15, 1916. The present corporation is the result of the.(5) <image>Dassault Aviation. Dassault Aviation S.A. () is a French manufacturer of military aircraft and business jets.It was founded in 1929 by Marcel Bloch as Société des Avions Marcel Bloch or \"MB\". After World War II, Marcel Bloch changed his name to Marcel Dassault, and the name of the company was changed to Avions Marcel Dassault on 20 January 1947.In 1971 Dassault acquired Breguet, forming Avions Marcel Dassault-Breguet Aviation (AMD-BA). In 1990 the company was renamed Dassault Aviation.The Dassault Aviation Group has been headed by Éric Trappier since 9 January 2013.## History.The Société des Avions Marcel Bloch was founded by Marcel Bloch in.\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/00/oven_00000127.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q114/Q1141409.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q66/Q66.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q66/Q66.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q66/Q66.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q460/Q460487.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 5,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q114/Q1141409.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q66/Q66.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q66/Q66.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q66/Q66.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q460/Q460487.jpg"
],
"candidate_texts": [
"Dassault Falcon 900. for dropping rescue stores.Falcon 900BRevised production version from 1991. Powered by 21.13 kN (4,750 lbf) TFE731-5BR-1C engines.Falcon 900CReplacement for 900B. Introduced in 2000.Long range version, with 22.24 kN (5,000 lbf) engines. This variant features TFE731-60 engines, with a range of 8,340 km (4,501 nm; 5,180 miles). Avionics by (Honeywell Primus). It entered service in 1996.Falcon 900EX EASyFalcon 900DXShorter-range production type. TFE731-60 engines.Falcon 900LXCurrent production variant of EX fitted with blended winglets. Range of 4750 (nmi).VC-900AItalian military designation for the 900EX.VC-900BItalian military designation for the 900EASy.## Operators.## Civil operators.A wide range of private owners, businesses and small airlines operate Falcon 900s.",
"Boeing. gas emissions was poised to grow, as air travel increases and ground vehicles use more alternative fuels like ethanol and biodiesel. Boeing estimates that biofuels could reduce flight-related greenhouse-gas emissions by 60 to 80%. The solution blends algae fuels with existing jet fuel.Boeing executives said the company was collaborating with Brazilian biofuels maker Tecbio, Aquaflow Bionomic of New Zealand, and other fuel developers around the world. As of 2007, Boeing had tested six fuels from these companies, and expected to test 20 fuels \"by the time we're done evaluating them\". Boeing also joined other aviation-related members in the Algal Biomass.",
"Boeing. 25% Hydro-treated Renewable Jet fuel and 25% of a Fischer–Tropsch fuel with successful results.## Electric propulsion.For NASA's N+3 future airliner program, Boeing has determined that hybrid electric engine technology is by far the best choice for its subsonic design. Hybrid electric propulsion has the potential to shorten takeoff distance and reduce noise. Boeing created a team to study electric propulsion in future generation of subsonic commercial aircraft. SUGAR for Subsonic Ultra Green Aircraft Research includes, BR&T, Boeing Commercial Airplanes, General Electric, and Georgia Tech. There are five main concepts the team is reviewing. SUGAR Free and Refined SUGAR, are two.",
"Boeing. The Boeing Company () is an American multinational corporation that designs, manufactures, and sells airplanes, rotorcraft, rockets, satellites, telecommunications equipment, and missiles worldwide. The company also provides leasing and product support services. Boeing is among the largest global aerospace manufacturers; it is the third-largest defense contractor in the world based on 2020 revenue, and is the largest exporter in the United States by dollar value. Boeing stock is included in the Dow Jones Industrial Average. Boeing is incorporated in Delaware.Boeing was founded by William Boeing in Seattle, Washington, on July 15, 1916. The present corporation is the result of the.",
"Dassault Aviation. Dassault Aviation S.A. () is a French manufacturer of military aircraft and business jets.It was founded in 1929 by Marcel Bloch as Société des Avions Marcel Bloch or \"MB\". After World War II, Marcel Bloch changed his name to Marcel Dassault, and the name of the company was changed to Avions Marcel Dassault on 20 January 1947.In 1971 Dassault acquired Breguet, forming Avions Marcel Dassault-Breguet Aviation (AMD-BA). In 1990 the company was renamed Dassault Aviation.The Dassault Aviation Group has been headed by Éric Trappier since 9 January 2013.## History.The Société des Avions Marcel Bloch was founded by Marcel Bloch in."
],
"ground_truth_position": 5,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/00/oven_00000127.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q114/Q1141409.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q66/Q66.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q66/Q66.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q66/Q66.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q460/Q460487.jpg"
],
"num_candidates": 5,
"qid": "5:142",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/00/oven_00000127.jpg",
"query_text": "What is the manufacturer of this aircraft?"
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Candidates:(1) A man riding an elephant through the water.(2) An elephant with white tusks walking through the water.(3) A man riding an elephant into the water.(4) A man is riding on the back of an elephant.(5) A man on an elephant moving on waters.\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000122998.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 1,
"style": "rule"
}
|
{
"candidate_images": [],
"candidate_texts": [
"A man riding an elephant through the water.",
"An elephant with white tusks walking through the water.",
"A man riding an elephant into the water.",
"A man is riding on the back of an elephant.",
"A man on an elephant moving on waters."
],
"ground_truth_position": 1,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000122998.jpg"
],
"num_candidates": 5,
"qid": "9:83452",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000122998.jpg",
"query_text": ""
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Candidates:(1) A couple of people riding surfboards on top of tall waves.(2) An image of a man going on the ocean waves.(3) A group of people riding surf boards on a wave.(4) A group of surfers are riding a wave.(5) Three surfers preparing to ride the waves at the beach.\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000127632.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 1,
"style": "rule"
}
|
{
"candidate_images": [],
"candidate_texts": [
"A couple of people riding surfboards on top of tall waves.",
"An image of a man going on the ocean waves.",
"A group of people riding surf boards on a wave.",
"A group of surfers are riding a wave.",
"Three surfers preparing to ride the waves at the beach."
],
"ground_truth_position": 1,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000127632.jpg"
],
"num_candidates": 5,
"qid": "9:188334",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000127632.jpg",
"query_text": ""
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Candidates:(1) <image>(2) <image>(3) <image>(4) <image>(5) <image>\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/ref/047/831.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/047/831_1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/078/290_0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/005/643_0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/014/006_1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/079/408_0.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 1,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/047/831_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/078/290_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/005/643_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/014/006_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/079/408_0.jpg"
],
"candidate_texts": [],
"ground_truth_position": 1,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/ref/047/831.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/047/831_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/078/290_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/005/643_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/014/006_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/079/408_0.jpg"
],
"num_candidates": 5,
"qid": "4:7641",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/ref/047/831.jpg",
"query_text": ""
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Candidates:(1) <image>(2) <image>(3) <image>(4) <image>(5) <image>\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/ref/011/600.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/061/929_1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/056/577_0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/058/586_0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/011/600_0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/087/271_0.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 4,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/061/929_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/056/577_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/058/586_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/011/600_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/087/271_0.jpg"
],
"candidate_texts": [],
"ground_truth_position": 4,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/ref/011/600.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/061/929_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/056/577_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/058/586_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/011/600_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/087/271_0.jpg"
],
"num_candidates": 5,
"qid": "4:1829",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/ref/011/600.jpg",
"query_text": ""
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Candidates:(1) <image>(2) <image>(3) <image>(4) <image>(5) <image>\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/ref/088/115.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/084/174_1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/088/115_0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/062/341_0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/090/500_1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/039/080_1.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 2,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/084/174_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/088/115_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/062/341_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/090/500_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/039/080_1.jpg"
],
"candidate_texts": [],
"ground_truth_position": 2,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/ref/088/115.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/084/174_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/088/115_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/062/341_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/090/500_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/039/080_1.jpg"
],
"num_candidates": 5,
"qid": "4:14053",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/ref/088/115.jpg",
"query_text": ""
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:Red floral print tie neck tunic.Candidates:(1) <image>(2) <image>(3) <image>(4) <image>(5) <image>\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/tops/blouses/61593288/61593288_3.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/tops/long_sleeved_tops/88068851/88068851_1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/tops/blouses/91152579/91152579_0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/tops/blouses/90172734/90172734_1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/tops/blouses/91212662/91212662_3.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 3,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/tops/blouses/61593288/61593288_3.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/tops/long_sleeved_tops/88068851/88068851_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/tops/blouses/91152579/91152579_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/tops/blouses/90172734/90172734_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/tops/blouses/91212662/91212662_3.jpg"
],
"candidate_texts": [],
"ground_truth_position": 3,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/tops/blouses/61593288/61593288_3.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/tops/long_sleeved_tops/88068851/88068851_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/tops/blouses/91152579/91152579_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/tops/blouses/90172734/90172734_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/tops/blouses/91212662/91212662_3.jpg"
],
"num_candidates": 5,
"qid": "1:8337",
"query_image": null,
"query_text": "Red floral print tie neck tunic."
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Candidates:(1) Two guys sitting on couch watching the two girls play the Wii.(2) A couple of women standing in front of a couch.(3) Two girls with wii remotes in a living room.(4) People playing on the wii while a man watches in the background.(5) Some people play with a Wii in their living room.\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000108201.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 1,
"style": "rule"
}
|
{
"candidate_images": [],
"candidate_texts": [
"Two guys sitting on couch watching the two girls play the Wii.",
"A couple of women standing in front of a couch.",
"Two girls with wii remotes in a living room.",
"People playing on the wii while a man watches in the background.",
"Some people play with a Wii in their living room."
],
"ground_truth_position": 1,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000108201.jpg"
],
"num_candidates": 5,
"qid": "9:1213",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000108201.jpg",
"query_text": ""
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Candidates:(1) <image>(2) <image>(3) <image>(4) <image>(5) <image>\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/ref/090/473.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/090/473_0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/029/267_0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/054/258_0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/001/270_1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/005/646_0.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 1,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/090/473_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/029/267_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/054/258_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/001/270_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/005/646_0.jpg"
],
"candidate_texts": [],
"ground_truth_position": 1,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/ref/090/473.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/090/473_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/029/267_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/054/258_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/001/270_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/005/646_0.jpg"
],
"num_candidates": 5,
"qid": "4:14415",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/ref/090/473.jpg",
"query_text": ""
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Which species of bird is this?Candidates:(1) Brown-headed nuthatch. The brown-headed nuthatch (\"Sitta pusilla\") is a small songbird found endemic to pine forests throughout the Southeastern United States. Genetic analyses indicated low differentiation between northern and southern populations in Florida, but the study also found lower genetic diversity among south Florida populations that may be a result of the increased habitat fragmentation that was documented. The Bahama nuthatch was formerly considered a subspecies (\"S. p. insularis\"), has since been reclassified as its own separate species. Two recent studies assessing vocalizations in Bahama and continental nuthatch populations found important differences. One of the studies also demonstrated that continental and Bahama.(2) Brown-headed nuthatch. The brown-headed nuthatch (\"Sitta pusilla\") is a small songbird found endemic to pine forests throughout the Southeastern United States. Genetic analyses indicated low differentiation between northern and southern populations in Florida, but the study also found lower genetic diversity among south Florida populations that may be a result of the increased habitat fragmentation that was documented. The Bahama nuthatch was formerly considered a subspecies (\"S. p. insularis\"), has since been reclassified as its own separate species. Two recent studies assessing vocalizations in Bahama and continental nuthatch populations found important differences. One of the studies also demonstrated that continental and Bahama.(3) Brown-headed nuthatch. are the world's smallest nuthatches. In the brown-headed nuthatch, the total length is 9 - 11 (cm), wingspan is 16 - 18 (cm) and body mass is 10 - 12 (g). This species sports a brown cap with narrow black eyeline and buff white cheeks, chin, and belly. Its wings are bluish-gray in color. A small white spot is found at the nape of the neck. The bird's call is a sharp \"whee-hyah\" sounding very similar to a \"rubber duck\" toy and particularly is loud for a bird its size. They also make softer \"pit pit pit\" calls while in.(4) Brown-headed nuthatch. such a cooperative groups where groups of 3-5 adults provide care at a single nest. Recent genetic assessments suggest some of the putatively non-breeding adults associated with these groups may actually breed with individuals in neighboring territories. This nuthatch also exhibits a wide range of other social behaviors that include social grooming and male-female duets similar to those observed for the Pygmy Nuthatch.The brown-headed nuthatch has been found to prefer making their nests at the top of short snags.Despite the other species' common name, the brown-headed nuthatch is about the same size as the pygmy nuthatch and the two species.(5) Brown-headed nuthatch. populations did not respond aggressively to calls of the other population. This type of call-response study is often used to help define cryptic species.The bird, like other nuthatches, possesses a sharp black nail-like beak, which it uses to pound open seeds. It is a frequent visitor to feeding stations and is highly fond of sunflower seeds and suet cakes. Bold and inquisitive, this bird is readily approachable by humans.The bird is regularly observed using a small chips of bark, small twigs, and pine needles held in its beak as tools to dig for insects. The nuthatch exhibits other curious behaviors.\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/00/oven_00874184.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 1,
"style": "rule"
}
|
{
"candidate_images": [],
"candidate_texts": [
"Brown-headed nuthatch. The brown-headed nuthatch (\"Sitta pusilla\") is a small songbird found endemic to pine forests throughout the Southeastern United States. Genetic analyses indicated low differentiation between northern and southern populations in Florida, but the study also found lower genetic diversity among south Florida populations that may be a result of the increased habitat fragmentation that was documented. The Bahama nuthatch was formerly considered a subspecies (\"S. p. insularis\"), has since been reclassified as its own separate species. Two recent studies assessing vocalizations in Bahama and continental nuthatch populations found important differences. One of the studies also demonstrated that continental and Bahama.",
"Brown-headed nuthatch. The brown-headed nuthatch (\"Sitta pusilla\") is a small songbird found endemic to pine forests throughout the Southeastern United States. Genetic analyses indicated low differentiation between northern and southern populations in Florida, but the study also found lower genetic diversity among south Florida populations that may be a result of the increased habitat fragmentation that was documented. The Bahama nuthatch was formerly considered a subspecies (\"S. p. insularis\"), has since been reclassified as its own separate species. Two recent studies assessing vocalizations in Bahama and continental nuthatch populations found important differences. One of the studies also demonstrated that continental and Bahama.",
"Brown-headed nuthatch. are the world's smallest nuthatches. In the brown-headed nuthatch, the total length is 9 - 11 (cm), wingspan is 16 - 18 (cm) and body mass is 10 - 12 (g). This species sports a brown cap with narrow black eyeline and buff white cheeks, chin, and belly. Its wings are bluish-gray in color. A small white spot is found at the nape of the neck. The bird's call is a sharp \"whee-hyah\" sounding very similar to a \"rubber duck\" toy and particularly is loud for a bird its size. They also make softer \"pit pit pit\" calls while in.",
"Brown-headed nuthatch. such a cooperative groups where groups of 3-5 adults provide care at a single nest. Recent genetic assessments suggest some of the putatively non-breeding adults associated with these groups may actually breed with individuals in neighboring territories. This nuthatch also exhibits a wide range of other social behaviors that include social grooming and male-female duets similar to those observed for the Pygmy Nuthatch.The brown-headed nuthatch has been found to prefer making their nests at the top of short snags.Despite the other species' common name, the brown-headed nuthatch is about the same size as the pygmy nuthatch and the two species.",
"Brown-headed nuthatch. populations did not respond aggressively to calls of the other population. This type of call-response study is often used to help define cryptic species.The bird, like other nuthatches, possesses a sharp black nail-like beak, which it uses to pound open seeds. It is a frequent visitor to feeding stations and is highly fond of sunflower seeds and suet cakes. Bold and inquisitive, this bird is readily approachable by humans.The bird is regularly observed using a small chips of bark, small twigs, and pine needles held in its beak as tools to dig for insects. The nuthatch exhibits other curious behaviors."
],
"ground_truth_position": 1,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/00/oven_00874184.jpg"
],
"num_candidates": 5,
"qid": "5:143922",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/00/oven_00874184.jpg",
"query_text": "Which species of bird is this?"
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:A river runs throughout the tree lined area.Candidates:(1) <image>(2) <image>(3) <image>(4) <image>(5) <image>\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000316928.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000504673.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000204423.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000276763.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000440217.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 1,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000316928.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000504673.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000204423.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000276763.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000440217.jpg"
],
"candidate_texts": [],
"ground_truth_position": 1,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000316928.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000504673.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000204423.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000276763.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000440217.jpg"
],
"num_candidates": 5,
"qid": "9:77953",
"query_image": null,
"query_text": "A river runs throughout the tree lined area."
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:Sometimes you really need to know whether it s going to rain.Candidates:(1) <image>(2) <image>(3) <image>(4) <image>(5) <image>\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/guardian/images/0210/065.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0705/368.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0560/991.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0666/754.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/washington_post/images/0419/196.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 5,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/guardian/images/0210/065.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0705/368.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0560/991.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0666/754.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/washington_post/images/0419/196.jpg"
],
"candidate_texts": [],
"ground_truth_position": 5,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/guardian/images/0210/065.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0705/368.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0560/991.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0666/754.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/washington_post/images/0419/196.jpg"
],
"num_candidates": 5,
"qid": "0:48413",
"query_image": null,
"query_text": "Sometimes you really need to know whether it s going to rain."
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Candidates:(1) Multicolor ruffle dress.(2) Red cocktail dress.(3) Purple crepe fit flare dress.(4) Red all day dress.(5) Blue ruffle trim sleeveless dress.\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/dresses/cocktail_dresses/90407403/90407403_0.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 2,
"style": "rule"
}
|
{
"candidate_images": [],
"candidate_texts": [
"Multicolor ruffle dress.",
"Red cocktail dress.",
"Purple crepe fit flare dress.",
"Red all day dress.",
"Blue ruffle trim sleeveless dress."
],
"ground_truth_position": 2,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/dresses/cocktail_dresses/90407403/90407403_0.jpg"
],
"num_candidates": 5,
"qid": "1:28373",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/dresses/cocktail_dresses/90407403/90407403_0.jpg",
"query_text": ""
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Candidates:(1) <image>(2) <image>(3) <image>(4) <image>(5) <image>\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/ref/054/403.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/075/493_0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/078/720_0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/002/356_1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/035/462_0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/054/403_0.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 5,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/075/493_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/078/720_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/002/356_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/035/462_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/054/403_0.jpg"
],
"candidate_texts": [],
"ground_truth_position": 5,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/ref/054/403.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/075/493_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/078/720_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/002/356_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/035/462_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/054/403_0.jpg"
],
"num_candidates": 5,
"qid": "4:8666",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/ref/054/403.jpg",
"query_text": ""
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:People have left flowers and piano instruction manuals outside the home of Ruthanne Lodato in Alexandria.Candidates:(1) <image>(2) <image>(3) <image>(4) <image>(5) <image>\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/washington_post/images/0115/411.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/washington_post/images/0354/111.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/bbc/images/0269/183.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0331/612.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/washington_post/images/0544/943.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 1,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/washington_post/images/0115/411.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/washington_post/images/0354/111.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/bbc/images/0269/183.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0331/612.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/washington_post/images/0544/943.jpg"
],
"candidate_texts": [],
"ground_truth_position": 1,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/washington_post/images/0115/411.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/washington_post/images/0354/111.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/bbc/images/0269/183.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0331/612.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/washington_post/images/0544/943.jpg"
],
"num_candidates": 5,
"qid": "0:2260",
"query_image": null,
"query_text": "People have left flowers and piano instruction manuals outside the home of Ruthanne Lodato in Alexandria."
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Make the pudding in large size on plates rather making into pieces.Candidates:(1) <image>(2) <image>(3) <image>(4) <image>(5) <image>\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/49/train-9286-1-img1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/92/train-11851-1-img0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/36/train-5830-2-img0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/65/train-4954-0-img0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/64/train-12083-0-img1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/80/train-187-3-img1.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 3,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/92/train-11851-1-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/36/train-5830-2-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/65/train-4954-0-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/64/train-12083-0-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/80/train-187-3-img1.jpg"
],
"candidate_texts": [],
"ground_truth_position": 3,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/49/train-9286-1-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/92/train-11851-1-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/36/train-5830-2-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/65/train-4954-0-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/64/train-12083-0-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/80/train-187-3-img1.jpg"
],
"num_candidates": 5,
"qid": "8:27248",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/49/train-9286-1-img1.jpg",
"query_text": "Make the pudding in large size on plates rather making into pieces."
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Show sleeping wolf.Candidates:(1) <image>(2) <image>(3) <image>(4) <image>(5) <image>\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/15/train-4528-2-img1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/71/train-8683-0-img1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/31/train-1163-0-img0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/61/train-10063-3-img1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/35/train-8578-1-img0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/75/train-359-3-img1.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 2,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/71/train-8683-0-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/31/train-1163-0-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/61/train-10063-3-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/35/train-8578-1-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/75/train-359-3-img1.jpg"
],
"candidate_texts": [],
"ground_truth_position": 2,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/15/train-4528-2-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/71/train-8683-0-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/31/train-1163-0-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/61/train-10063-3-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/35/train-8578-1-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/75/train-359-3-img1.jpg"
],
"num_candidates": 5,
"qid": "8:21739",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/15/train-4528-2-img1.jpg",
"query_text": "Show sleeping wolf."
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:Blue two-tone denim jacket.Candidates:(1) <image>(2) <image>(3) <image>(4) <image>(5) <image>\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/jackets/denim_jackets/90800918/90800918_4.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/jackets/denim_jackets/91158545/91158545_0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/jackets/denim_jackets/65577790/65577790_2.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/jackets/denim_jackets/91194644/91194644_3.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/jackets/denim_jackets/65221559/65221559_9.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 2,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/jackets/denim_jackets/90800918/90800918_4.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/jackets/denim_jackets/91158545/91158545_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/jackets/denim_jackets/65577790/65577790_2.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/jackets/denim_jackets/91194644/91194644_3.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/jackets/denim_jackets/65221559/65221559_9.jpg"
],
"candidate_texts": [],
"ground_truth_position": 2,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/jackets/denim_jackets/90800918/90800918_4.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/jackets/denim_jackets/91158545/91158545_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/jackets/denim_jackets/65577790/65577790_2.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/jackets/denim_jackets/91194644/91194644_3.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/jackets/denim_jackets/65221559/65221559_9.jpg"
],
"num_candidates": 5,
"qid": "1:13708",
"query_image": null,
"query_text": "Blue two-tone denim jacket."
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Candidates:(1) <image>(2) <image>(3) <image>(4) <image>(5) <image>\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/ref/028/710.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/028/710_0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/076/574_0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/041/894_0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/060/957_1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/085/768_1.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 1,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/028/710_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/076/574_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/041/894_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/060/957_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/085/768_1.jpg"
],
"candidate_texts": [],
"ground_truth_position": 1,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/ref/028/710.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/028/710_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/076/574_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/041/894_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/060/957_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/085/768_1.jpg"
],
"num_candidates": 5,
"qid": "4:4582",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/ref/028/710.jpg",
"query_text": ""
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Which government has executive power of this sport?Candidates:(1) <image>Ice hockey. amateur leagues leading to national championships were in place. One of the first was the Swiss National League A, founded in 1916. Today, professional leagues have been introduced in most countries of Europe. Top European leagues include the Kontinental Hockey League, the Czech Extraliga, the Finnish Liiga and the Swedish Hockey League.## Game.While the general characteristics of the game remain constant, the exact rules depend on the particular code of play being used. The two most important codes are those of the IIHF and the NHL. Both of these codes, and others, originated from Canadian rules of ice hockey of.(2) <image>Ice hockey. of whom is the goaltender. Ice hockey is a full contact sport.Ice hockey is one of the sports featured in the Winter Olympics while its premiere international amateur competition, the IIHF World Championships, are governed by the International Ice Hockey Federation (IIHF) for both men's and women's competitions. Ice hockey is also played as a professional sport.In North America as well as many European countries, the sport is known simply as hockey. However, in other countries, \"hockey\" usually refers to field hockey. In 1994 hockey was officially recognized as Canada's national winter sport.The modern sport of ice hockey was developed.(3) <image>Ice hockey. outside of the rink, because the puck is considered dead once it leaves the perimeter of the rink. The referee may also blow the whistle for a stoppage in play if the puck is jammed along the boards when 2 or more players are battling for the puck for a long time, or if the puck is stuck on the back of any of the two nets for a period of time.Under IIHF rules, each team may carry a maximum of 20 players and two goaltenders on their roster. NHL rules restrict the total number of players per game to.(4) <image>Ice hockey. the International Ice Hockey Federation.As the popularity of ice hockey as a spectator sport grew, earlier rinks were replaced by larger rinks. Most of the early indoor ice rinks have been demolished; Montreal's Victoria Rink, built in 1862, was demolished in 1925. Many older rinks succumbed to fire, such as Denman Arena, Dey's Arena, Quebec Skating Rink and Montreal Arena, a hazard of the buildings' wood construction. The Stannus Street Rink in Windsor, Nova Scotia (built in 1897) may be the oldest still in existence; however, it is no longer used for hockey. The Aberdeen Pavilion (built in 1898) in.(5) <image>Ice hockey. (NHL). In the early 1900s, the Canadian rules were adopted by the Ligue Internationale de Hockey Sur Glace, in Paris, France, the precursor of the International Ice Hockey Federation. The sport was played for the first time at the Olympics during the 1920 Summer Olympics.While women also played during the game's early formative years, it was not until organizers began to officially remove body checking from female ice hockey beginning in the mid-1980s that it began to gain greater popularity, which by then had spread to Europe and a variety of other countries. The first IIHF Women's World Championship was.\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/01/oven_01069796.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q414/Q41466.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q414/Q41466.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q414/Q41466.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q414/Q41466.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q414/Q41466.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 1,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q414/Q41466.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q414/Q41466.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q414/Q41466.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q414/Q41466.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q414/Q41466.jpg"
],
"candidate_texts": [
"Ice hockey. amateur leagues leading to national championships were in place. One of the first was the Swiss National League A, founded in 1916. Today, professional leagues have been introduced in most countries of Europe. Top European leagues include the Kontinental Hockey League, the Czech Extraliga, the Finnish Liiga and the Swedish Hockey League.## Game.While the general characteristics of the game remain constant, the exact rules depend on the particular code of play being used. The two most important codes are those of the IIHF and the NHL. Both of these codes, and others, originated from Canadian rules of ice hockey of.",
"Ice hockey. of whom is the goaltender. Ice hockey is a full contact sport.Ice hockey is one of the sports featured in the Winter Olympics while its premiere international amateur competition, the IIHF World Championships, are governed by the International Ice Hockey Federation (IIHF) for both men's and women's competitions. Ice hockey is also played as a professional sport.In North America as well as many European countries, the sport is known simply as hockey. However, in other countries, \"hockey\" usually refers to field hockey. In 1994 hockey was officially recognized as Canada's national winter sport.The modern sport of ice hockey was developed.",
"Ice hockey. outside of the rink, because the puck is considered dead once it leaves the perimeter of the rink. The referee may also blow the whistle for a stoppage in play if the puck is jammed along the boards when 2 or more players are battling for the puck for a long time, or if the puck is stuck on the back of any of the two nets for a period of time.Under IIHF rules, each team may carry a maximum of 20 players and two goaltenders on their roster. NHL rules restrict the total number of players per game to.",
"Ice hockey. the International Ice Hockey Federation.As the popularity of ice hockey as a spectator sport grew, earlier rinks were replaced by larger rinks. Most of the early indoor ice rinks have been demolished; Montreal's Victoria Rink, built in 1862, was demolished in 1925. Many older rinks succumbed to fire, such as Denman Arena, Dey's Arena, Quebec Skating Rink and Montreal Arena, a hazard of the buildings' wood construction. The Stannus Street Rink in Windsor, Nova Scotia (built in 1897) may be the oldest still in existence; however, it is no longer used for hockey. The Aberdeen Pavilion (built in 1898) in.",
"Ice hockey. (NHL). In the early 1900s, the Canadian rules were adopted by the Ligue Internationale de Hockey Sur Glace, in Paris, France, the precursor of the International Ice Hockey Federation. The sport was played for the first time at the Olympics during the 1920 Summer Olympics.While women also played during the game's early formative years, it was not until organizers began to officially remove body checking from female ice hockey beginning in the mid-1980s that it began to gain greater popularity, which by then had spread to Europe and a variety of other countries. The first IIHF Women's World Championship was."
],
"ground_truth_position": 1,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/01/oven_01069796.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q414/Q41466.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q414/Q41466.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q414/Q41466.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q414/Q41466.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q414/Q41466.jpg"
],
"num_candidates": 5,
"qid": "6:147203",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/01/oven_01069796.jpg",
"query_text": "Which government has executive power of this sport?"
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Shows a team of similar dogs (all the same color) pulling a sled with two people on it.Candidates:(1) <image>(2) <image>(3) <image>(4) <image>(5) <image>\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/61/train-10884-1-img0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/26/train-5301-3-img0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/95/train-7227-2-img1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/83/train-1119-2-img1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/89/train-6686-3-img0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/82/train-12306-2-img1.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 1,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/26/train-5301-3-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/95/train-7227-2-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/83/train-1119-2-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/89/train-6686-3-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/82/train-12306-2-img1.jpg"
],
"candidate_texts": [],
"ground_truth_position": 1,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/61/train-10884-1-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/26/train-5301-3-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/95/train-7227-2-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/83/train-1119-2-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/89/train-6686-3-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/82/train-12306-2-img1.jpg"
],
"num_candidates": 5,
"qid": "8:17602",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/61/train-10884-1-img0.jpg",
"query_text": "Shows a team of similar dogs (all the same color) pulling a sled with two people on it."
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Candidates:(1) <image>(2) <image>(3) <image>(4) <image>(5) <image>\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/ref/078/648.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/064/719_0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/031/937_1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/078/648_0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/061/652_0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/097/560_1.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 3,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/064/719_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/031/937_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/078/648_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/061/652_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/097/560_1.jpg"
],
"candidate_texts": [],
"ground_truth_position": 3,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/ref/078/648.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/064/719_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/031/937_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/078/648_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/061/652_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/097/560_1.jpg"
],
"num_candidates": 5,
"qid": "4:12517",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/ref/078/648.jpg",
"query_text": ""
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Candidates:(1) <image>(2) <image>(3) <image>(4) <image>(5) <image>\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/ref/048/719.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/001/280_0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/021/708_0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/011/083_0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/048/719_1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/002/547_1.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 4,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/001/280_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/021/708_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/011/083_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/048/719_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/002/547_1.jpg"
],
"candidate_texts": [],
"ground_truth_position": 4,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/ref/048/719.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/001/280_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/021/708_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/011/083_0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/048/719_1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/distort/002/547_1.jpg"
],
"num_candidates": 5,
"qid": "4:7779",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/nights_images/ref/048/719.jpg",
"query_text": ""
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:European Commission President Jose Manuel Barroso left talks to United Nations SecretaryGeneral Ban Kimoon after a joint press conference during the World Economic Forum in Davos Switzerland Thursday Jan 23 2014.Candidates:(1) <image>(2) <image>(3) <image>(4) <image>(5) <image>\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/washington_post/images/0468/848.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/bbc/images/0395/171.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0158/333.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/guardian/images/0259/861.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/washington_post/images/0045/243.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 3,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/washington_post/images/0468/848.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/bbc/images/0395/171.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0158/333.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/guardian/images/0259/861.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/washington_post/images/0045/243.jpg"
],
"candidate_texts": [],
"ground_truth_position": 3,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/washington_post/images/0468/848.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/bbc/images/0395/171.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0158/333.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/guardian/images/0259/861.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/washington_post/images/0045/243.jpg"
],
"num_candidates": 5,
"qid": "0:157498",
"query_image": null,
"query_text": "European Commission President Jose Manuel Barroso left talks to United Nations SecretaryGeneral Ban Kimoon after a joint press conference during the World Economic Forum in Davos Switzerland Thursday Jan 23 2014."
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Shows the side view of another dog of the same breed standing in a cement area.Candidates:(1) <image>(2) <image>(3) <image>(4) <image>(5) <image>\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/9/train-1720-1-img1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/16/train-6755-3-img1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/6/train-6234-3-img0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/83/train-7768-2-img1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/57/train-148-3-img1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/69/train-2718-3-img1.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 4,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/16/train-6755-3-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/6/train-6234-3-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/83/train-7768-2-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/57/train-148-3-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/69/train-2718-3-img1.jpg"
],
"candidate_texts": [],
"ground_truth_position": 4,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/9/train-1720-1-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/16/train-6755-3-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/6/train-6234-3-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/83/train-7768-2-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/57/train-148-3-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/69/train-2718-3-img1.jpg"
],
"num_candidates": 5,
"qid": "8:21972",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/9/train-1720-1-img1.jpg",
"query_text": "Shows the side view of another dog of the same breed standing in a cement area."
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Candidates:(1) Multicolor soft slub short sleeve tee w/ soft rib trim.(2) White superfine v neck tee.(3) Black long sleeve crew tee.(4) White must-have easy pocket tee.(5) White short sleeve back pleat tee.\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/tops/short_sleeve_tops/91039078/91039078_4.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 1,
"style": "rule"
}
|
{
"candidate_images": [],
"candidate_texts": [
"Multicolor soft slub short sleeve tee w/ soft rib trim.",
"White superfine v neck tee.",
"Black long sleeve crew tee.",
"White must-have easy pocket tee.",
"White short sleeve back pleat tee."
],
"ground_truth_position": 1,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/tops/short_sleeve_tops/91039078/91039078_4.jpg"
],
"num_candidates": 5,
"qid": "1:13352",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/tops/short_sleeve_tops/91039078/91039078_4.jpg",
"query_text": ""
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:Two stuffed animals sit together on a chair.Candidates:(1) <image>(2) <image>(3) <image>(4) <image>(5) <image>\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000475567.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000528033.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000235843.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000051012.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000045836.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 2,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000475567.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000528033.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000235843.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000051012.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000045836.jpg"
],
"candidate_texts": [],
"ground_truth_position": 2,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000475567.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000528033.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000235843.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000051012.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000045836.jpg"
],
"num_candidates": 5,
"qid": "9:86799",
"query_image": null,
"query_text": "Two stuffed animals sit together on a chair."
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:Did king edward I visit the Banff castle the same century it was built?Candidates:(1) The castle was visited by King Edward I of England in 1296 and also in 1298 after defeating William Wallace at the Battle of Falkirk during the Wars of Scottish Independence. The castle was garrisoned with English troops until being captured by the Scots in 1310.(2) Banff Castle is a ruined former royal castle near Banff, Aberdeenshire, Scotland. Built as a motte and bailey castle in the 12th century and held by the Comyns, Earl of Buchan.(3) King Henry II, who added on to the castle, in 1155 confirmed William d'Aubigny as Earl of Arundel, with the honour and the castle of Arundel.(4) Geoffrey died attacking Burwell Castle the following year, leaving Cambridge Castle secure. Under Henry II the castle was adequately maintained, but little additional work was undertaken to improve it.(5) Some Scots nobles refused to swear fealty to Bruce, and were disinherited. They left Scotland to join forces with Edward Balliol, son of King John Balliol (r. 1292–1296), whom Edward I had deposed in 1296.\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[] |
multimodal_reranking
|
{
"ground_truth": 1,
"style": "rule"
}
|
{
"candidate_images": [],
"candidate_texts": [
"The castle was visited by King Edward I of England in 1296 and also in 1298 after defeating William Wallace at the Battle of Falkirk during the Wars of Scottish Independence. The castle was garrisoned with English troops until being captured by the Scots in 1310.",
"Banff Castle is a ruined former royal castle near Banff, Aberdeenshire, Scotland. Built as a motte and bailey castle in the 12th century and held by the Comyns, Earl of Buchan.",
"King Henry II, who added on to the castle, in 1155 confirmed William d'Aubigny as Earl of Arundel, with the honour and the castle of Arundel.",
"Geoffrey died attacking Burwell Castle the following year, leaving Cambridge Castle secure. Under Henry II the castle was adequately maintained, but little additional work was undertaken to improve it.",
"Some Scots nobles refused to swear fealty to Bruce, and were disinherited. They left Scotland to join forces with Edward Balliol, son of King John Balliol (r. 1292–1296), whom Edward I had deposed in 1296."
],
"ground_truth_position": 1,
"images": [],
"num_candidates": 5,
"qid": "2:24401",
"query_image": null,
"query_text": "Did king edward I visit the Banff castle the same century it was built?"
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Which species of fungi is this?Candidates:(1) Cladonia cristatella. Cladonia cristatella, commonly known as the British soldiers lichen or the British soldiers cup lichen, is a fruticose, cup lichen belonging to the family Cladoniaceae. The species was first described scientifically by American botanist Edward Tuckerman in 1858.List of \"Cladonia\" species.(2) Phoradendron californicum. Phoradendron californicum, the desert mistletoe or mesquite mistletoe, is a hemiparasitic plant native to southern California, Nevada, Arizona, Sonora, Sinaloa and Baja California. It can be found in the Mojave and Sonoran Deserts at elevations of up to 1400 m (4600 feet).The mistletoe is a leafless plant that attaches to host plants, often leguminous woody desert trees such as \"Cercidium\" and \"Prosopis\". Desert mistletoe takes water and minerals from its host plants but it does its own photosynthesis, making it a hemiparasite.During the winter it produces inconspicuous, fragrant flowers. Female desert mistletoe plants produce red to clear berries that are.(3) Cladonia cristatella. Cladonia cristatella, commonly known as the British soldiers lichen or the British soldiers cup lichen, is a fruticose, cup lichen belonging to the family Cladoniaceae. The species was first described scientifically by American botanist Edward Tuckerman in 1858.List of \"Cladonia\" species.(4) Lobaria pulmonaria. by the Hesquiaht in British Columbia, Canada. An ethnophytotherapeutical survey of the high Molise region in central-southern Italy revealed that \"L. pulmonaria\" is used as an antiseptic, and is rubbed on wounds.A hot-water extract prepared using this species has been shown to have anti-inflammatory and ulcer-preventing activities. Also, methanol extracts were shown to have a protective effect on the gastrointestinal system of rats, possibly by reducing oxidative stress and reducing the inflammatory effects of neutrophils. Furthermore, methanol extracts also have potent antioxidative activity and reducing power, probably due to the presence of phenolic compounds.## Other uses.\"Lobaria pulmonaria\" has also been.(5) Ramalina menziesii. Ramalina menziesii, the lace lichen or fishnet, is a pale yellowish-green to grayish-green foliose lichen. It grows up to a meter long, hanging from bark and twigs in a distinctive net-like or lace-like pattern that is unlike any other lichen in North America. It becomes a deeper green when wet. Apothecia are lecanorine. it is an important food source for deer in the Coast Range of California, and a source of nest material for birds. It is highly variable in its growth form, with branches sometimes so slender as to appear like strands, sometimes tiny, and sometimes large with broadly.\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/00/oven_00925214.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 1,
"style": "rule"
}
|
{
"candidate_images": [],
"candidate_texts": [
"Cladonia cristatella. Cladonia cristatella, commonly known as the British soldiers lichen or the British soldiers cup lichen, is a fruticose, cup lichen belonging to the family Cladoniaceae. The species was first described scientifically by American botanist Edward Tuckerman in 1858.List of \"Cladonia\" species.",
"Phoradendron californicum. Phoradendron californicum, the desert mistletoe or mesquite mistletoe, is a hemiparasitic plant native to southern California, Nevada, Arizona, Sonora, Sinaloa and Baja California. It can be found in the Mojave and Sonoran Deserts at elevations of up to 1400 m (4600 feet).The mistletoe is a leafless plant that attaches to host plants, often leguminous woody desert trees such as \"Cercidium\" and \"Prosopis\". Desert mistletoe takes water and minerals from its host plants but it does its own photosynthesis, making it a hemiparasite.During the winter it produces inconspicuous, fragrant flowers. Female desert mistletoe plants produce red to clear berries that are.",
"Cladonia cristatella. Cladonia cristatella, commonly known as the British soldiers lichen or the British soldiers cup lichen, is a fruticose, cup lichen belonging to the family Cladoniaceae. The species was first described scientifically by American botanist Edward Tuckerman in 1858.List of \"Cladonia\" species.",
"Lobaria pulmonaria. by the Hesquiaht in British Columbia, Canada. An ethnophytotherapeutical survey of the high Molise region in central-southern Italy revealed that \"L. pulmonaria\" is used as an antiseptic, and is rubbed on wounds.A hot-water extract prepared using this species has been shown to have anti-inflammatory and ulcer-preventing activities. Also, methanol extracts were shown to have a protective effect on the gastrointestinal system of rats, possibly by reducing oxidative stress and reducing the inflammatory effects of neutrophils. Furthermore, methanol extracts also have potent antioxidative activity and reducing power, probably due to the presence of phenolic compounds.## Other uses.\"Lobaria pulmonaria\" has also been.",
"Ramalina menziesii. Ramalina menziesii, the lace lichen or fishnet, is a pale yellowish-green to grayish-green foliose lichen. It grows up to a meter long, hanging from bark and twigs in a distinctive net-like or lace-like pattern that is unlike any other lichen in North America. It becomes a deeper green when wet. Apothecia are lecanorine. it is an important food source for deer in the Coast Range of California, and a source of nest material for birds. It is highly variable in its growth form, with branches sometimes so slender as to appear like strands, sometimes tiny, and sometimes large with broadly."
],
"ground_truth_position": 1,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/00/oven_00925214.jpg"
],
"num_candidates": 5,
"qid": "5:147719",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/00/oven_00925214.jpg",
"query_text": "Which species of fungi is this?"
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:A room with a couch sitting on a wooden floor.Candidates:(1) <image>(2) <image>(3) <image>(4) <image>(5) <image>\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000355908.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000458410.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000453972.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000444207.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000581821.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 3,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000355908.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000458410.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000453972.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000444207.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000581821.jpg"
],
"candidate_texts": [],
"ground_truth_position": 3,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000355908.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000458410.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000453972.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000444207.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000581821.jpg"
],
"num_candidates": 5,
"qid": "9:75850",
"query_image": null,
"query_text": "A room with a couch sitting on a wooden floor."
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:Does the train station at Earl's Court Road in London have one train track only?Candidates:(1) <image>Barons Court tube station in London, spring 2013 (1) Barons Court tube station in West Kensington in the London Borough of Hammersmith and Fulham, London, England, Great Britain.(2) <image>Baker Street stn Circle line look west Baker Street tube station Circle/Hammersmith & City line platforms looking west/anti-clockwise.(3) <image>Baker Street tube station MMB 05 A London Underground A Stock Metropolitan Line unit stands at Baker Street.(4) <image>Clapton Railway Station - geograph.org.uk - 1768642 Clapton Railway Station.(5) <image>London Train Station.\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30125078.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30304412.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30046844.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30084453.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30178913.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 5,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30125078.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30304412.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30046844.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30084453.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30178913.jpg"
],
"candidate_texts": [
"Barons Court tube station in London, spring 2013 (1) Barons Court tube station in West Kensington in the London Borough of Hammersmith and Fulham, London, England, Great Britain.",
"Baker Street stn Circle line look west Baker Street tube station Circle/Hammersmith & City line platforms looking west/anti-clockwise.",
"Baker Street tube station MMB 05 A London Underground A Stock Metropolitan Line unit stands at Baker Street.",
"Clapton Railway Station - geograph.org.uk - 1768642 Clapton Railway Station.",
"London Train Station."
],
"ground_truth_position": 5,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30125078.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30304412.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30046844.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30084453.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30178913.jpg"
],
"num_candidates": 5,
"qid": "2:12309",
"query_image": null,
"query_text": "Does the train station at Earl's Court Road in London have one train track only?"
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Who found this aircraft?Candidates:(1) <image>Fairchild Aircraft. was a merge of Fairchild-Hiller Corporation, division and subsidiaries: Fairchild Aircraft Marketing Company, Fairchild Aircraft Services Division, Fairchild Republic Division, Fairchild Space and Electronics Division, Fairchild Stratos Division, Burns Aero Seat Company, Inc., Fairchild Arms International, Ltd., Fairchild Aviation (Asia) Ltd., Fairchild Aviation (Holland) N.V., Fairchild-Germantown Development Company, Inc. and S.J. Industries, Inc. Before 1971, Fairchild Industries was a term used to include many of the companies of its founder Sherman Mills Fairchild.After the name change, the company purchased Swearingen and manufactured the Fairchild Swearingen Metroliner, a successful commuter aircraft that gained orders from the U.S. military as the C-26.(2) <image>Cessna. purchased the Beechcraft and Hawker Aircraft corporations, Cessna ceased operations as a subsidiary company and joined the others as one of the three distinct brands produced by Textron Aviation.Throughout its history, and especially after World War II, Cessna became best-known for producing high-wing small aircraft with the wings themselves placed above the main body, a structural feature that remains an iconic distinction among planes still being manufactured under the Cessna brand. ## History.## Origins.Clyde Cessna, a farmer in Rago, Kansas, built his own aircraft and flew it in June 1911. He was the first person to do so between the.(3) <image>Tabasco sauce. sumptuous dinner parties. Mr. White's table no doubt groaned with the region's varied fare—drawing inspiration from European, Caribbean, and Cajun sources—but one of his favorite sauces was of his own devising, made from a pepper named for its origins in the Mexican state of Tabasco. White added it to various dishes and bottled it for his guests. Although the McIlhennys have tried to dismiss the possibility, it seems clear now that in 1849, a full two decades before Edmund McIlhenny professed to discover the Tabasco pepper, White was already growing Tabasco chilies on his plantation.\"Rothfeder cited January 26, 1850, letter.(4) <image>Cessna. Mississippi River and the Rocky Mountains. Cessna started his wood-and-fabric aircraft ventures in Enid, Oklahoma, testing many of his early planes on the salt flats. When bankers in Enid refused to lend him more money to build his planes, he moved to Wichita.Cessna Aircraft was formed when Clyde Cessna and Victor Roos became partners in the Cessna-Roos Aircraft Company in 1927. Roos resigned just one month into the partnership, selling back his interest to Cessna. Shortly afterward, Roos's name was dropped from the company name.The Cessna DC-6 earned certification on the same day as the stock market crash of 1929,.(5) <image>Fairchild Aircraft. a twin-engine trainer, was manufactured at a former rayon mill in Burlington, North Carolina. Also large numbers of the Fairchild 24 (C-61/Argus) were produced for the military (principally as the Argus for the Royal Air Force), and continued production after the war for the civilian market. Fairchild ranked 73rd among United States corporations in the value of World War II military production contracts.## Postwar.The C-82 Packet led to the C-119 Flying Boxcar, another U.S. military transport aircraft. The C-119 could carry cargo, personnel, stretcher patients and mechanized equipment with the ability to make \"paradrops\" of cargo and troops. The first.\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/00/oven_00003672.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q113/Q1136766.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q308/Q308889.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q335/Q335016.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q308/Q308889.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q113/Q1136766.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 2,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q113/Q1136766.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q308/Q308889.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q335/Q335016.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q308/Q308889.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q113/Q1136766.jpg"
],
"candidate_texts": [
"Fairchild Aircraft. was a merge of Fairchild-Hiller Corporation, division and subsidiaries: Fairchild Aircraft Marketing Company, Fairchild Aircraft Services Division, Fairchild Republic Division, Fairchild Space and Electronics Division, Fairchild Stratos Division, Burns Aero Seat Company, Inc., Fairchild Arms International, Ltd., Fairchild Aviation (Asia) Ltd., Fairchild Aviation (Holland) N.V., Fairchild-Germantown Development Company, Inc. and S.J. Industries, Inc. Before 1971, Fairchild Industries was a term used to include many of the companies of its founder Sherman Mills Fairchild.After the name change, the company purchased Swearingen and manufactured the Fairchild Swearingen Metroliner, a successful commuter aircraft that gained orders from the U.S. military as the C-26.",
"Cessna. purchased the Beechcraft and Hawker Aircraft corporations, Cessna ceased operations as a subsidiary company and joined the others as one of the three distinct brands produced by Textron Aviation.Throughout its history, and especially after World War II, Cessna became best-known for producing high-wing small aircraft with the wings themselves placed above the main body, a structural feature that remains an iconic distinction among planes still being manufactured under the Cessna brand. ## History.## Origins.Clyde Cessna, a farmer in Rago, Kansas, built his own aircraft and flew it in June 1911. He was the first person to do so between the.",
"Tabasco sauce. sumptuous dinner parties. Mr. White's table no doubt groaned with the region's varied fare—drawing inspiration from European, Caribbean, and Cajun sources—but one of his favorite sauces was of his own devising, made from a pepper named for its origins in the Mexican state of Tabasco. White added it to various dishes and bottled it for his guests. Although the McIlhennys have tried to dismiss the possibility, it seems clear now that in 1849, a full two decades before Edmund McIlhenny professed to discover the Tabasco pepper, White was already growing Tabasco chilies on his plantation.\"Rothfeder cited January 26, 1850, letter.",
"Cessna. Mississippi River and the Rocky Mountains. Cessna started his wood-and-fabric aircraft ventures in Enid, Oklahoma, testing many of his early planes on the salt flats. When bankers in Enid refused to lend him more money to build his planes, he moved to Wichita.Cessna Aircraft was formed when Clyde Cessna and Victor Roos became partners in the Cessna-Roos Aircraft Company in 1927. Roos resigned just one month into the partnership, selling back his interest to Cessna. Shortly afterward, Roos's name was dropped from the company name.The Cessna DC-6 earned certification on the same day as the stock market crash of 1929,.",
"Fairchild Aircraft. a twin-engine trainer, was manufactured at a former rayon mill in Burlington, North Carolina. Also large numbers of the Fairchild 24 (C-61/Argus) were produced for the military (principally as the Argus for the Royal Air Force), and continued production after the war for the civilian market. Fairchild ranked 73rd among United States corporations in the value of World War II military production contracts.## Postwar.The C-82 Packet led to the C-119 Flying Boxcar, another U.S. military transport aircraft. The C-119 could carry cargo, personnel, stretcher patients and mechanized equipment with the ability to make \"paradrops\" of cargo and troops. The first."
],
"ground_truth_position": 2,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/00/oven_00003672.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q113/Q1136766.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q308/Q308889.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q335/Q335016.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q308/Q308889.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q113/Q1136766.jpg"
],
"num_candidates": 5,
"qid": "6:141638",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/00/oven_00003672.jpg",
"query_text": "Who found this aircraft?"
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>What is the sea level in metre of this mountain?Candidates:(1) <image>Longmen Grottoes. Republic of China in 1949, the grottoes have been declared as protected area and are being conserved. The Constitution of China, under Article 22, which among other issues also provides for protection of the natural heritage sites, has been further defined under various legal instruments enacted to protect and conserve this cultural heritage of China.The Longmen Relics Care Agency was established in 1953 under the Ministry of Culture. A 1954 site inventory was undertaken by the newly established Longmen Caves Cultural Relics Management and Conservation Office. The State Council declared the Longmen Grottoes as a national cultural monument needing special.(2) <image>Mount Wutai. Mount Wutai, also known by its Chinese name Wutaishan and as is a sacred Buddhist site at the headwaters of the Qingshui in Shanxi Province, China. Its central area is surrounded by a cluster of flat-topped peaks roughly corresponding to the cardinal directions. The north peak (Beitai Ding or Yedou Feng) is the highest (3061 (m)) and is also the highest point in northern China.As host to over 53 sacred monasteries, Mount Wutai is home to many of China's most important monasteries and temples. It was inscribed as a UNESCO World Heritage Site in 2009 and named a AAAAA tourist.(3) <image>Monserrate. Monserrate (named after Catalan homonym mountain \"Montserrat\") is a high mountain over 10,000 feet high that dominates the city center of Bogotá, the capital city of Colombia. It rises to 3152 (meters) above the sea level, where there is a church (built in the 17th century) with a shrine, devoted to \"El Señor Caído\" (\"The Fallen Lord\").The Mountain, already considered sacred in pre-Columbian times when the area was inhabited by the indigenous Muisca, is a pilgrim destination, as well as a major tourist attraction. In addition to the church, the summit contains restaurants, cafeteria, souvenir shops and many smaller tourist.(4) <image>Longmen Grottoes. Tang Dynasty who lived in Luoyang during his later years. The tomb is located on the hill top. It is approached from the west bank after crossing a bridge across the Yi River. The tomb is a circular mound of earth of 4 meters height with a circumference of 52 meters. The tomb is 2.80 meters high and has the poet's name inscribed on it as Bai Juyi.## Preservation and restoration.One of the major achievements in this direction is the recognition given to the monuments by UNESCO in declaring the Longmen Grottoes a heritage monument, after due evaluation over a.(5) <image>Mount Wutai. Mount Wutai, also known by its Chinese name Wutaishan and as is a sacred Buddhist site at the headwaters of the Qingshui in Shanxi Province, China. Its central area is surrounded by a cluster of flat-topped peaks roughly corresponding to the cardinal directions. The north peak (Beitai Ding or Yedou Feng) is the highest (3061 (m)) and is also the highest point in northern China.As host to over 53 sacred monasteries, Mount Wutai is home to many of China's most important monasteries and temples. It was inscribed as a UNESCO World Heritage Site in 2009 and named a AAAAA tourist.\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/00/oven_00026589.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q273/Q273834.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q120/Q120314.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q152/Q152073.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q273/Q273834.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q120/Q120314.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 2,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q273/Q273834.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q120/Q120314.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q152/Q152073.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q273/Q273834.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q120/Q120314.jpg"
],
"candidate_texts": [
"Longmen Grottoes. Republic of China in 1949, the grottoes have been declared as protected area and are being conserved. The Constitution of China, under Article 22, which among other issues also provides for protection of the natural heritage sites, has been further defined under various legal instruments enacted to protect and conserve this cultural heritage of China.The Longmen Relics Care Agency was established in 1953 under the Ministry of Culture. A 1954 site inventory was undertaken by the newly established Longmen Caves Cultural Relics Management and Conservation Office. The State Council declared the Longmen Grottoes as a national cultural monument needing special.",
"Mount Wutai. Mount Wutai, also known by its Chinese name Wutaishan and as is a sacred Buddhist site at the headwaters of the Qingshui in Shanxi Province, China. Its central area is surrounded by a cluster of flat-topped peaks roughly corresponding to the cardinal directions. The north peak (Beitai Ding or Yedou Feng) is the highest (3061 (m)) and is also the highest point in northern China.As host to over 53 sacred monasteries, Mount Wutai is home to many of China's most important monasteries and temples. It was inscribed as a UNESCO World Heritage Site in 2009 and named a AAAAA tourist.",
"Monserrate. Monserrate (named after Catalan homonym mountain \"Montserrat\") is a high mountain over 10,000 feet high that dominates the city center of Bogotá, the capital city of Colombia. It rises to 3152 (meters) above the sea level, where there is a church (built in the 17th century) with a shrine, devoted to \"El Señor Caído\" (\"The Fallen Lord\").The Mountain, already considered sacred in pre-Columbian times when the area was inhabited by the indigenous Muisca, is a pilgrim destination, as well as a major tourist attraction. In addition to the church, the summit contains restaurants, cafeteria, souvenir shops and many smaller tourist.",
"Longmen Grottoes. Tang Dynasty who lived in Luoyang during his later years. The tomb is located on the hill top. It is approached from the west bank after crossing a bridge across the Yi River. The tomb is a circular mound of earth of 4 meters height with a circumference of 52 meters. The tomb is 2.80 meters high and has the poet's name inscribed on it as Bai Juyi.## Preservation and restoration.One of the major achievements in this direction is the recognition given to the monuments by UNESCO in declaring the Longmen Grottoes a heritage monument, after due evaluation over a.",
"Mount Wutai. Mount Wutai, also known by its Chinese name Wutaishan and as is a sacred Buddhist site at the headwaters of the Qingshui in Shanxi Province, China. Its central area is surrounded by a cluster of flat-topped peaks roughly corresponding to the cardinal directions. The north peak (Beitai Ding or Yedou Feng) is the highest (3061 (m)) and is also the highest point in northern China.As host to over 53 sacred monasteries, Mount Wutai is home to many of China's most important monasteries and temples. It was inscribed as a UNESCO World Heritage Site in 2009 and named a AAAAA tourist."
],
"ground_truth_position": 2,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/00/oven_00026589.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q273/Q273834.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q120/Q120314.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q152/Q152073.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q273/Q273834.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/wikipedia_images_full/Q120/Q120314.jpg"
],
"num_candidates": 5,
"qid": "6:160360",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/oven_images/00/oven_00026589.jpg",
"query_text": "What is the sea level in metre of this mountain?"
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Candidates:(1) A person on a surfboard is surfing on a crashing wave.(2) A surfer is riding on a surfboard against a wave.(3) A person in a black wetsuit surfing on a wave alone.(4) A single person riding a big wave with a surfboard.(5) The surfers is surfing against a high wave.\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000297595.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 1,
"style": "rule"
}
|
{
"candidate_images": [],
"candidate_texts": [
"A person on a surfboard is surfing on a crashing wave.",
"A surfer is riding on a surfboard against a wave.",
"A person in a black wetsuit surfing on a wave alone.",
"A single person riding a big wave with a surfboard.",
"The surfers is surfing against a high wave."
],
"ground_truth_position": 1,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000297595.jpg"
],
"num_candidates": 5,
"qid": "9:147998",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000297595.jpg",
"query_text": ""
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:A person standing outside of a building with an umbrella.Candidates:(1) <image>(2) <image>(3) <image>(4) <image>(5) <image>\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000081594.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000061729.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000451352.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000281602.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000285106.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 3,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000081594.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000061729.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000451352.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000281602.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000285106.jpg"
],
"candidate_texts": [],
"ground_truth_position": 3,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000081594.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000061729.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/train2014/COCO_train2014_000000451352.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000281602.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000285106.jpg"
],
"num_candidates": 5,
"qid": "9:6588",
"query_image": null,
"query_text": "A person standing outside of a building with an umbrella."
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Only two animals stirring up dust.Candidates:(1) <image>(2) <image>(3) <image>(4) <image>(5) <image>\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/47/train-12032-1-img0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/30/train-363-3-img1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/15/train-4325-3-img1.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/16/train-10008-1-img0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/66/train-125-1-img0.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/55/train-10100-2-img1.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 1,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/30/train-363-3-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/15/train-4325-3-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/16/train-10008-1-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/66/train-125-1-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/55/train-10100-2-img1.jpg"
],
"candidate_texts": [],
"ground_truth_position": 1,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/47/train-12032-1-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/30/train-363-3-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/15/train-4325-3-img1.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/16/train-10008-1-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/66/train-125-1-img0.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/55/train-10100-2-img1.jpg"
],
"num_candidates": 5,
"qid": "8:23253",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/cirr_images/train/47/train-12032-1-img0.jpg",
"query_text": "Only two animals stirring up dust."
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Candidates:(1) There is plenty of cheese and other types of vegetables on a plate.(2) A plate of food on a table.(3) A white plate silverware bowl and cup with different types of foods.(4) A whit plate topped with vegetables and cheese.(5) A plat of vegetables, and cheese and a bowl of nuts and milk on a table.\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000288313.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 1,
"style": "rule"
}
|
{
"candidate_images": [],
"candidate_texts": [
"There is plenty of cheese and other types of vegetables on a plate.",
"A plate of food on a table.",
"A white plate silverware bowl and cup with different types of foods.",
"A whit plate topped with vegetables and cheese.",
"A plat of vegetables, and cheese and a bowl of nuts and milk on a table."
],
"ground_truth_position": 1,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000288313.jpg"
],
"num_candidates": 5,
"qid": "9:121324",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/mscoco_images/val2014/COCO_val2014_000000288313.jpg",
"query_text": ""
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:Which direction do the tips of the ears of a Japanese Serow point?Candidates:(1) <image>Hemitragus jemlahicus2.(2) <image>Sambar walk Sambar (Cervus unicolor) in Ranthambore National Park.(3) <image>Cervus albirostris male male.(4) <image>Cervus nippon yakushimae 04-2 Sika Deer.(5) <image>Japanese serow - Berlin zoo Japanese serow (Capricornis crispus) found in the Berlin zoo.\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30370065.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30070697.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30087762.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30343825.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30152750.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 5,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30370065.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30070697.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30087762.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30343825.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30152750.jpg"
],
"candidate_texts": [
"Hemitragus jemlahicus2.",
"Sambar walk Sambar (Cervus unicolor) in Ranthambore National Park.",
"Cervus albirostris male male.",
"Cervus nippon yakushimae 04-2 Sika Deer.",
"Japanese serow - Berlin zoo Japanese serow (Capricornis crispus) found in the Berlin zoo."
],
"ground_truth_position": 5,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30370065.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30070697.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30087762.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30343825.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30152750.jpg"
],
"num_candidates": 5,
"qid": "2:12764",
"query_image": null,
"query_text": "Which direction do the tips of the ears of a Japanese Serow point?"
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:<image>Candidates:(1) Black luxe dress.(2) Multicolor full length dress.(3) Black draped cold shoulder dress.(4) Black more cold shoulder dress.(5) Multicolor long dress black solid.\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/dresses/maxi_and_long_dresses/90775505/90775505_2.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 4,
"style": "rule"
}
|
{
"candidate_images": [],
"candidate_texts": [
"Black luxe dress.",
"Multicolor full length dress.",
"Black draped cold shoulder dress.",
"Black more cold shoulder dress.",
"Multicolor long dress black solid."
],
"ground_truth_position": 4,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/dresses/maxi_and_long_dresses/90775505/90775505_2.jpg"
],
"num_candidates": 5,
"qid": "1:12350",
"query_image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/fashion200k_images/dresses/maxi_and_long_dresses/90775505/90775505_2.jpg",
"query_text": ""
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:What is the main primary color of Cerasela Patrascu's warm up outfit for the 2010 Olympics.Candidates:(1) <image>2017-02-05 Olena Shkhumova (Teamstaffel) by Sandro Halank–1.(2) <image>Women's 1500m, 2014 Winter Olympics, Ireen Wüst (2) Women's 1500m, 2014 Winter Olympics, Ireen Wust.(3) <image>CeraselaPatrascu Cerasela Patrascu during 2010 World Artistic Gymnastics Championships.(4) <image>2010 Winter Olympics - Curling - Women - Draw 5.a.(5) <image>2011 TEB Free 279 Jessica Dubé Sébastien Wolfe.\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30116545.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30313247.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30187729.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30049196.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30231941.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 3,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30116545.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30313247.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30187729.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30049196.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30231941.jpg"
],
"candidate_texts": [
"2017-02-05 Olena Shkhumova (Teamstaffel) by Sandro Halank–1.",
"Women's 1500m, 2014 Winter Olympics, Ireen Wüst (2) Women's 1500m, 2014 Winter Olympics, Ireen Wust.",
"CeraselaPatrascu Cerasela Patrascu during 2010 World Artistic Gymnastics Championships.",
"2010 Winter Olympics - Curling - Women - Draw 5.a.",
"2011 TEB Free 279 Jessica Dubé Sébastien Wolfe."
],
"ground_truth_position": 3,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30116545.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30313247.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30187729.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30049196.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/webqa_images/30231941.jpg"
],
"num_candidates": 5,
"qid": "2:9913",
"query_image": null,
"query_text": "What is the main primary color of Cerasela Patrascu's warm up outfit for the 2010 Olympics."
}
|
lamra_rerank
|
[
{
"content": "You are a helpful assistant for multimodal retrieval tasks.\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>\n{\"type\": \"function\", \"function\": {\"name\": \"crop_image\", \"description\": \"Zoom in on the image based on the bounding box coordinates.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"bbox_2d\": {\"type\": \"array\", \"description\": \"coordinates for bounding box of the area you want to zoom in. minimum value is 0 and maximum value is the width/height of the image.\", \"items\": {\"type\": \"number\"}}, \"target_image\": {\"type\": \"number\", \"description\": \"The index of the image to crop. Index from 1 to the number of images. Choose 1 to operate on original image.\"}}, \"required\": [\"bbox_2d\", \"target_image\"]}}\n{\"type\": \"function\", \"function\": {\"name\": \"select_images\", \"description\": \"Select exactly one candidate image for detailed analysis from the provided candidate list.\", \"parameters\": {\"type\": \"object\", \"properties\": {\"target_images\": {\"type\": \"array\", \"description\": \"List containing exactly one candidate image index to select for detailed analysis (e.g., [3] to select candidate 3).\", \"items\": {\"type\": \"integer\", \"description\": \"Candidate image index from 1 to the number of candidates.\"}, \"minItems\": 1, \"maxItems\": 1}}, \"required\": [\"target_images\"]}}}\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>",
"role": "system"
},
{
"content": "Query:A family photograph of Alexander Panin top left in his third grade class In an Oct 28 letter Panin told his mother he was nt sure when he started to think about dedicating himself to transforming life through technology.Candidates:(1) <image>(2) <image>(3) <image>(4) <image>(5) <image>\n\nGuidelines: You are given a query (which can be text, image, or both) and multiple candidate items (which can be text, image, or both). Your task is to evaluate each candidate against the query and select the best match. Determine if it is beneficial to employ the given visual operations (tools). You can use the following tools:\n1. `crop_image`: Zoom in on specific areas of the query image or selected candidate images for detailed analysis\n2. `select_images`: Select specific candidate images from the provided list for detailed comparison\n\nAnalyze the visual and textual features step by step and provide your final answer in the format:\n<think>Your reasoning process here</think><answer>Directly answer with the most matching number</answer>",
"role": "user"
}
] |
[
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0587/179.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0471/525.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0042/933.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0529/316.jpg"
},
{
"image": "/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0428/520.jpg"
}
] |
multimodal_reranking
|
{
"ground_truth": 4,
"style": "rule"
}
|
{
"candidate_images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0587/179.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0471/525.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0042/933.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0529/316.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0428/520.jpg"
],
"candidate_texts": [],
"ground_truth_position": 4,
"images": [
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0587/179.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0471/525.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0042/933.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0529/316.jpg",
"/remote-home1/hxzhuang/cdy/LamRA/data/mbeir_images/visualnews_images/usa_today/images/0428/520.jpg"
],
"num_candidates": 5,
"qid": "0:191534",
"query_image": null,
"query_text": "A family photograph of Alexander Panin top left in his third grade class In an Oct 28 letter Panin told his mother he was nt sure when he started to think about dedicating himself to transforming life through technology."
}
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.