user
string | videos
list | gt
int64 | predict
string | pred_order
list | top1_correct
bool | gt_pos
int64 | latency_ms
float64 | index
int64 | request_messages
list |
|---|---|---|---|---|---|---|---|---|---|
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A video of scissor kick jumping at track is shown. A few people take turns at varying heights and all are successful. Others are running for other events in the background."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_ZOQSDsJYXIA.mp4",
"activitynet/videos/v_AJ15GW-sS5M.mp4",
"activitynet/videos/v_l3EBfLkfAX8.mp4",
"activitynet/videos/v_ps0a-GGomX4.mp4",
"activitynet/videos/v_qlqF8K072UU.mp4"
] | 3
|
<answer>3</answer>
|
[
3
] | true
| 1
| 12,035.01186
| 200
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A video of scissor kick jumping at track is shown. A few people take turns at varying heights and all are successful. Others are running for other events in the background.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_ZOQSDsJYXIA.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_AJ15GW-sS5M.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_l3EBfLkfAX8.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_ps0a-GGomX4.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_qlqF8K072UU.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A person is seen sitting on a train with skateboard sitting in their laps and another skateboarder sleeping. The people walk for a bit and then are seen riding their boards along a city. The men ride along the city while several people on bikes ride around and the camera pans over people in the city. More shots of the city are shown as well as a large group of people indoors and outdoors."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_BDQHEemWnSk.mp4",
"activitynet/videos/v_4T8uFygBeNo.mkv",
"activitynet/videos/v_2ooY3GqZieg.mkv",
"activitynet/videos/v_0AbJgWxIYVI.mp4",
"activitynet/videos/v_qsYElirHVUU.mp4"
] | 3
|
<answer>3</answer>
|
[
3
] | true
| 1
| 24,287.368816
| 201
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A person is seen sitting on a train with skateboard sitting in their laps and another skateboarder sleeping. The people walk for a bit and then are seen riding their boards along a city. The men ride along the city while several people on bikes ride around and the camera pans over people in the city. More shots of the city are shown as well as a large group of people indoors and outdoors.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_BDQHEemWnSk.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_4T8uFygBeNo.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_2ooY3GqZieg.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_0AbJgWxIYVI.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_qsYElirHVUU.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "men and women are standing in a dock playing drums. people in a marching band are standing on an open area playing cymbals and drummers. people are standing in the background watching the matching band."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_Qci4EFEIZuo.mp4",
"activitynet/videos/v_0KTued0g034.mp4",
"activitynet/videos/v_WOZbWqJMkRg.mp4",
"activitynet/videos/v_mqwC7rqeXsk.mp4",
"activitynet/videos/v_MSSb3wPd5hM.mp4"
] | 5
|
<answer>5</answer>
|
[
5
] | true
| 1
| 18,638.5932
| 202
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"men and women are standing in a dock playing drums. people in a marching band are standing on an open area playing cymbals and drummers. people are standing in the background watching the matching band.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_Qci4EFEIZuo.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_0KTued0g034.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_WOZbWqJMkRg.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_mqwC7rqeXsk.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_MSSb3wPd5hM.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A woman is seen looking at the camera and making various movements while several camera effects are added. The girl runs product through her hair and leads into her brushing her hair and spraying in more product. She continues to comb and ends by pointing to the camera and showing off her hair."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_5SpWmZxECqc.mkv",
"activitynet/videos/v_osjru9UsWsI.mp4",
"activitynet/videos/v_ibWb6iRQiD4.mp4",
"activitynet/videos/v_7O9kkDxEvaY.mp4",
"activitynet/videos/v_0h4UT-2XTAw.mp4"
] | 1
|
<answer>2</answer>
|
[
2
] | false
| 0
| 19,315.975596
| 203
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A woman is seen looking at the camera and making various movements while several camera effects are added. The girl runs product through her hair and leads into her brushing her hair and spraying in more product. She continues to comb and ends by pointing to the camera and showing off her hair.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_5SpWmZxECqc.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_osjru9UsWsI.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_ibWb6iRQiD4.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_7O9kkDxEvaY.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_0h4UT-2XTAw.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A man is seen kneeling down before a car tires and pointing to objects. He then puts a block down next to the tire as well as taking off the hubcap. The man puts a jack down and unscrews the tire taking it off. He screws in the new one and puts the hubcap back on."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_GKBYgS99oiI.mp4",
"activitynet/videos/v_3TLhUYQ8geM.mp4",
"activitynet/videos/v_Ey2SmPzJTKM.mp4",
"activitynet/videos/v_furUOKw0Qzs.mp4",
"activitynet/videos/v_-doxoUNGLJE.mp4"
] | 4
|
<answer>1</answer>
|
[
1
] | false
| 0
| 18,019.924858
| 205
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A man is seen kneeling down before a car tires and pointing to objects. He then puts a block down next to the tire as well as taking off the hubcap. The man puts a jack down and unscrews the tire taking it off. He screws in the new one and puts the hubcap back on.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_GKBYgS99oiI.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_3TLhUYQ8geM.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_Ey2SmPzJTKM.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_furUOKw0Qzs.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_-doxoUNGLJE.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A man is seen speaking to the camera and begins pouring ice into a glass followed by various alcohols. He mixes more ingredients into the glass and finishes by putting a straw, cherry, and lemon in the glass, presenting it to the camera."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_hoYF0DhYVOI.mp4",
"activitynet/videos/v_rzm4V_McRhQ.mp4",
"activitynet/videos/v_PxGggNnMGtQ.mkv",
"activitynet/videos/v_rXwSSTGmvb8.mp4",
"activitynet/videos/v_KzxVQ19pRUU.mp4"
] | 3
|
<answer>2</answer>
|
[
2
] | false
| 0
| 21,323.682373
| 204
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A man is seen speaking to the camera and begins pouring ice into a glass followed by various alcohols. He mixes more ingredients into the glass and finishes by putting a straw, cherry, and lemon in the glass, presenting it to the camera.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_hoYF0DhYVOI.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_rzm4V_McRhQ.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_PxGggNnMGtQ.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_rXwSSTGmvb8.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_KzxVQ19pRUU.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A woman is seated on a work table and holds a paint brush. The woman paints a picture of long stems and leaves of a plant. The woman paints flower petals onto the painting."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_4ImpZRtbzYw.mp4",
"activitynet/videos/v_I9NukwdINyY.mp4",
"activitynet/videos/v_zy7rd78yBnY.mp4",
"activitynet/videos/v_tj0sI8M3tro.mkv",
"activitynet/videos/v__3xMhj4mbsk.mp4"
] | 3
|
<answer>3</answer>
|
[
3
] | true
| 1
| 23,567.985265
| 206
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A woman is seated on a work table and holds a paint brush. The woman paints a picture of long stems and leaves of a plant. The woman paints flower petals onto the painting.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_4ImpZRtbzYw.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_I9NukwdINyY.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_zy7rd78yBnY.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_tj0sI8M3tro.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v__3xMhj4mbsk.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "Two men are sitting at the table with their arms positioned to arm wrestle. The young men are holding each other's hands as they arm wrestle each other. The man in white beat the man in black sleeveless shirt."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_qwxmpiaT-kk.mp4",
"activitynet/videos/v_jeaaS1NK_d4.mp4",
"activitynet/videos/v_JXMD8Obk0yg.mp4",
"activitynet/videos/v_1NAlbF88oUI.mp4",
"activitynet/videos/v_a_uamUiKq1o.mp4"
] | 1
|
<answer>1</answer>
|
[
1
] | true
| 1
| 18,615.815231
| 207
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"Two men are sitting at the table with their arms positioned to arm wrestle. The young men are holding each other's hands as they arm wrestle each other. The man in white beat the man in black sleeveless shirt.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_qwxmpiaT-kk.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_jeaaS1NK_d4.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_JXMD8Obk0yg.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_1NAlbF88oUI.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_a_uamUiKq1o.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "An older man wearing glasses and headphones is seen playing a set of bongo drums while closes his eyes to concentrate. The man plays the instrument continuously and finishes by putting one hand up."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_EKfhRuD3x9s.mp4",
"activitynet/videos/v_maHLwXvNN3w.mp4",
"activitynet/videos/v_UuJwtJBJ7oU.mp4",
"activitynet/videos/v_trl-RCWyhb0.mp4",
"activitynet/videos/v_EbGq9gXcXLQ.mp4"
] | 4
|
<answer>4</answer>
|
[
4
] | true
| 1
| 20,571.371563
| 208
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"An older man wearing glasses and headphones is seen playing a set of bongo drums while closes his eyes to concentrate. The man plays the instrument continuously and finishes by putting one hand up.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_EKfhRuD3x9s.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_maHLwXvNN3w.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_UuJwtJBJ7oU.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_trl-RCWyhb0.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_EbGq9gXcXLQ.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A person takes a key off a wall , gasses and starts a tractor. A boy in a hat pretends to be a girl. The first boy rides the tractor while rapping. The boy almost hits a tree in the tractor. The boy in orange rides past a kneeling boy. The boy is doing home work and stands on the tractor. We see a game screen and toy tractor. The third boy starts rapping and riding the tractor mower. The third boy pretends to sleep on the tractor. The three boys dance around the tractor. All three boys pause and rap together. We see a person in a window."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_ILF-93buuSY.mp4",
"activitynet/videos/v_puK4NxGKNdQ.mp4",
"activitynet/videos/v_vGcH8N8sJlM.mp4",
"activitynet/videos/v_bDwGZOk7njI.mp4",
"activitynet/videos/v_l4C-l6XeNRc.mp4"
] | 5
|
<answer>5</answer>
|
[
5
] | true
| 1
| 15,570.075832
| 209
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A person takes a key off a wall , gasses and starts a tractor. A boy in a hat pretends to be a girl. The first boy rides the tractor while rapping. The boy almost hits a tree in the tractor. The boy in orange rides past a kneeling boy. The boy is doing home work and stands on the tractor. We see a game screen and toy tractor. The third boy starts rapping and riding the tractor mower. The third boy pretends to sleep on the tractor. The three boys dance around the tractor. All three boys pause and rap together. We see a person in a window.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_ILF-93buuSY.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_puK4NxGKNdQ.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_vGcH8N8sJlM.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_bDwGZOk7njI.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_l4C-l6XeNRc.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "Scenes of various pole vaulters vaulting before audiences are shown. A slow motion video of one of the vaults is shown. Another pole vaulter's vault is shown. The vault is repeated in slow motion. A still image of the vaulter kneeling next to a scoreboard is shown."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_MxKuqpxmKKk.mkv",
"activitynet/videos/v_U0qUFAPUg_Y.mp4",
"activitynet/videos/v_FrkXeG1YoKg.mp4",
"activitynet/videos/v_JoiZmVQCLCI.mkv",
"activitynet/videos/v_XZZRyOhxQBE.mp4"
] | 3
|
<answer>2</answer>
|
[
2
] | false
| 0
| 9,320.177526
| 210
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"Scenes of various pole vaulters vaulting before audiences are shown. A slow motion video of one of the vaults is shown. Another pole vaulter's vault is shown. The vault is repeated in slow motion. A still image of the vaulter kneeling next to a scoreboard is shown.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_MxKuqpxmKKk.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_U0qUFAPUg_Y.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_FrkXeG1YoKg.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_JoiZmVQCLCI.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_XZZRyOhxQBE.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "Three people are seen standing around when one woman does hop scotch followed by another man doing hop scotch. One last man hopes his way down and the woman takes one more turn."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_s2VpBgSWIPg.mp4",
"activitynet/videos/v_q8-iXvYyCGg.mp4",
"activitynet/videos/v_z85nM9V4058.mp4",
"activitynet/videos/v_CR_79ZjQG_w.mp4",
"activitynet/videos/v_iDMzTPfELoc.mp4"
] | 2
|
<answer>1</answer>
|
[
1
] | false
| 0
| 10,605.34587
| 212
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"Three people are seen standing around when one woman does hop scotch followed by another man doing hop scotch. One last man hopes his way down and the woman takes one more turn.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_s2VpBgSWIPg.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_q8-iXvYyCGg.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_z85nM9V4058.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_CR_79ZjQG_w.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_iDMzTPfELoc.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "There are several men dressed in biking gear and helmets participating in a dirt bike competition. There are several spectators watching the event. They are driving at high speed on a circular race track. The bikers take a huge leap and jump over the obstacle and land on the other side of the race track."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_ATk8OkvNHHQ.mp4",
"activitynet/videos/v_aTQaYDmcMDY.mp4",
"activitynet/videos/v_gXdFGYPKClE.mp4",
"activitynet/videos/v_fmRio4-6Xqc.mp4",
"activitynet/videos/v_2aw1pVJsnKs.mp4"
] | 4
|
<answer>3</answer>
|
[
3
] | false
| 0
| 21,895.378611
| 211
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"There are several men dressed in biking gear and helmets participating in a dirt bike competition. There are several spectators watching the event. They are driving at high speed on a circular race track. The bikers take a huge leap and jump over the obstacle and land on the other side of the race track.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_ATk8OkvNHHQ.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_aTQaYDmcMDY.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_gXdFGYPKClE.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_fmRio4-6Xqc.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_2aw1pVJsnKs.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A group of men in multi colored turbans form two groups at an outdoor event and play a game of tug of war. A group of men dressed in blue shirts and blue shorts form a line next to a rope stretched out on the ground. Another group of men forms across from them taking the same position next to the rope on the ground. The men begin to play tug of war in front of a large group of onlookers on either side."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_I3BWhaDRxGk.mp4",
"activitynet/videos/v_wKThOOUV6lY.mp4",
"activitynet/videos/v_mFWRIp164r4.mp4",
"activitynet/videos/v_9L-aeZsgwZs.mp4",
"activitynet/videos/v_XELYXH6fqeA.mp4"
] | 3
|
<answer>2</answer>
|
[
2
] | false
| 0
| 18,127.841018
| 214
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A group of men in multi colored turbans form two groups at an outdoor event and play a game of tug of war. A group of men dressed in blue shirts and blue shorts form a line next to a rope stretched out on the ground. Another group of men forms across from them taking the same position next to the rope on the ground. The men begin to play tug of war in front of a large group of onlookers on either side.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_I3BWhaDRxGk.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_wKThOOUV6lY.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_mFWRIp164r4.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_9L-aeZsgwZs.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_XELYXH6fqeA.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A large group of women are standing in a dance room with walled mirrors, wooden floors and red words appear on the screen that say "En Sus Marcas Listos Fuera Daddy Yankee". One woman wearing a pink and black top jumps to the front of the group near the mirror and she quickly begins to dance as the women behind her follow her every move as best as they can. A new group of women are shown and the same woman is leading but she's now wearing a yellow and black top as she does her routine and when she's done the screen fades to white."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_DmAOCYuMgtE.mp4",
"activitynet/videos/v_sxQ9H3c5bRM.mp4",
"activitynet/videos/v_MWnYL4JiMP0.mp4",
"activitynet/videos/v_pouxwDABDrg.mp4",
"activitynet/videos/v_GhXniQgRUTY.mp4"
] | 3
|
<answer>3</answer>
|
[
3
] | true
| 1
| 25,033.81528
| 213
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A large group of women are standing in a dance room with walled mirrors, wooden floors and red words appear on the screen that say \"En Sus Marcas Listos Fuera Daddy Yankee\". One woman wearing a pink and black top jumps to the front of the group near the mirror and she quickly begins to dance as the women behind her follow her every move as best as they can. A new group of women are shown and the same woman is leading but she's now wearing a yellow and black top as she does her routine and when she's done the screen fades to white.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_DmAOCYuMgtE.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_sxQ9H3c5bRM.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_MWnYL4JiMP0.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_pouxwDABDrg.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_GhXniQgRUTY.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A woman leads an exercise class in an elaborate dance routine. A woman in a green tank top stands in front of a group of other women and leads them in an elaborate dance routine. The woman completes dances straight into the camera, at a profile and also uses elaborate arm and leg movements spanning the distance of the floor in the routine."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_gT_8511vwVE.mp4",
"activitynet/videos/v_pouxwDABDrg.mp4",
"activitynet/videos/v_vL8Hy6lcnF8.mp4",
"activitynet/videos/v_MC0L0ljTUiw.mp4",
"activitynet/videos/v_DmAOCYuMgtE.mp4"
] | 5
|
<answer>1</answer>
|
[
1
] | false
| 0
| 18,362.022391
| 215
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A woman leads an exercise class in an elaborate dance routine. A woman in a green tank top stands in front of a group of other women and leads them in an elaborate dance routine. The woman completes dances straight into the camera, at a profile and also uses elaborate arm and leg movements spanning the distance of the floor in the routine.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_gT_8511vwVE.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_pouxwDABDrg.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_vL8Hy6lcnF8.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_MC0L0ljTUiw.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_DmAOCYuMgtE.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A very thick man walks slowly into a a large room filled with people and he stands in front of a barbell with red and blue weights on both ends. The man takes a few breaths then reaches down and puts his hands on the barbell. After a while the man lifts the barbell to he neck, then jumps and bends his legs at the same time putting the barbell above his head, then he throws the barbell to the ground."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_70bS0DkAeDo.mp4",
"activitynet/videos/v_U9Dcet1qdRE.mp4",
"activitynet/videos/v_AyicWbHhUWc.mp4",
"activitynet/videos/v_sGFbsMKkoYs.mp4",
"activitynet/videos/v_rRkwB9EcEMs.mp4"
] | 3
|
<answer>1</answer>
|
[
1
] | false
| 0
| 18,586.543429
| 216
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A very thick man walks slowly into a a large room filled with people and he stands in front of a barbell with red and blue weights on both ends. The man takes a few breaths then reaches down and puts his hands on the barbell. After a while the man lifts the barbell to he neck, then jumps and bends his legs at the same time putting the barbell above his head, then he throws the barbell to the ground.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_70bS0DkAeDo.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_U9Dcet1qdRE.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_AyicWbHhUWc.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_sGFbsMKkoYs.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_rRkwB9EcEMs.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A photo album with the words "Blow Drying My Hair" appear. A woman shares the steps to properly blow dry your hair, including using tension, and demonstrates. She finishes by checking and comparing the length of her hair to previous years."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_4SLvbRa2NI0.mp4",
"activitynet/videos/v_dfjl7sS1IGo.mp4",
"activitynet/videos/v_N2WxAkVh-C4.mp4",
"activitynet/videos/v_p1fpQ4yR1co.mp4",
"activitynet/videos/v_4ZoBfU4b5Ko.mp4"
] | 2
|
<answer>2</answer>
|
[
2
] | true
| 1
| 21,736.762684
| 217
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A photo album with the words \"Blow Drying My Hair\" appear. A woman shares the steps to properly blow dry your hair, including using tension, and demonstrates. She finishes by checking and comparing the length of her hair to previous years.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_4SLvbRa2NI0.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_dfjl7sS1IGo.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_N2WxAkVh-C4.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_p1fpQ4yR1co.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_4ZoBfU4b5Ko.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A woman is hula hooping in front of a swimming pool. The hula hoops flies away and she walks towards the camera."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_w1FFMG52FZE.mp4",
"activitynet/videos/v_hT_4wWPNYxo.mp4",
"activitynet/videos/v_EhjiQFHfDmY.mp4",
"activitynet/videos/v_EpV0Zmg50nQ.mp4",
"activitynet/videos/v_VvlJjaLwGqY.mp4"
] | 1
|
<answer>1</answer>
|
[
1
] | true
| 1
| 16,982.024761
| 218
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A woman is hula hooping in front of a swimming pool. The hula hoops flies away and she walks towards the camera.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_w1FFMG52FZE.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_hT_4wWPNYxo.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_EhjiQFHfDmY.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_EpV0Zmg50nQ.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_VvlJjaLwGqY.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A player is seen standing ready with another running and throwing the ball. The ball is hit and another catches the ball and throws it back. Several more shots of people playing are shown as well a team members watching and laughing on the side. A player shakes hands with another and shows a slow motion recap of the previous play. The players all huddle up and speak and shows one walking away followed by several more."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_35SpLMcN-m8.mp4",
"activitynet/videos/v_7X3wPRKuAsU.mp4",
"activitynet/videos/v_3f6G-qzwzfg.mp4",
"activitynet/videos/v_QBXswoKU4S4.mp4",
"activitynet/videos/v_mkEME_iWi9o.mp4"
] | 4
|
<answer>5</answer>
|
[
5
] | false
| 0
| 10,674.450537
| 220
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A player is seen standing ready with another running and throwing the ball. The ball is hit and another catches the ball and throws it back. Several more shots of people playing are shown as well a team members watching and laughing on the side. A player shakes hands with another and shows a slow motion recap of the previous play. The players all huddle up and speak and shows one walking away followed by several more.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_35SpLMcN-m8.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_7X3wPRKuAsU.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_3f6G-qzwzfg.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_QBXswoKU4S4.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_mkEME_iWi9o.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A woman is talking as she is seated in a room. She shows off a braid in her hair. She then demonstrates how to make the braid."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_41__Qick6tM.mp4",
"activitynet/videos/v_3La7NPOBVN8.mp4",
"activitynet/videos/v_FNlEHAIh6LQ.mp4",
"activitynet/videos/v_sVma83g_wmg.mp4",
"activitynet/videos/v_ZLmoqxkCJL0.mp4"
] | 2
|
<answer>2</answer>
|
[
2
] | true
| 1
| 22,287.346732
| 219
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A woman is talking as she is seated in a room. She shows off a braid in her hair. She then demonstrates how to make the braid.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_41__Qick6tM.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_3La7NPOBVN8.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_FNlEHAIh6LQ.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_sVma83g_wmg.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_ZLmoqxkCJL0.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "An athlete holds a heavy ball, then he spins around and throw the ball. A person measure the distance or the heavy ball throw."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_J2gGPC98yec.mp4",
"activitynet/videos/v_lAN2pe1lW-o.mp4",
"activitynet/videos/v_j6HDZh7W6Z4.mp4",
"activitynet/videos/v_LAZHNzFbDNY.mp4",
"activitynet/videos/v_Bule85koN3o.mp4"
] | 5
|
<answer>5</answer>
|
[
5
] | true
| 1
| 18,783.150463
| 221
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"An athlete holds a heavy ball, then he spins around and throw the ball. A person measure the distance or the heavy ball throw.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_J2gGPC98yec.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_lAN2pe1lW-o.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_j6HDZh7W6Z4.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_LAZHNzFbDNY.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_Bule85koN3o.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A close up is shown of an old star wars poster. A toy is shown, also from the movie. A young man begins playing the saxophone into a microphone. When he finishes, he walks away."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_cdufbM2OCwM.mkv",
"activitynet/videos/v_FDLhpMkJwCM.mkv",
"activitynet/videos/v_a1nRXQZ6-Fo.mp4",
"activitynet/videos/v_aN9vCyXMbb8.mp4",
"activitynet/videos/v_wdD-UHM8rTg.mp4"
] | 1
|
<answer>4</answer>
|
[
4
] | false
| 0
| 14,423.351835
| 222
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A close up is shown of an old star wars poster. A toy is shown, also from the movie. A young man begins playing the saxophone into a microphone. When he finishes, he walks away.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_cdufbM2OCwM.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_FDLhpMkJwCM.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_a1nRXQZ6-Fo.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_aN9vCyXMbb8.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_wdD-UHM8rTg.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "The little girl is putting the spoon on the other side of the sink. The little girl is brushing the big wooden spoon and put it under the running water and put the wooden spoon on the other side of the sink. The girl picked up the bottle, rub the sponge on it, then rinse it with water, and put it on the strainer. The girl picked up the plate and rub the sponge on plate."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_u8ykXBc2Efs.mkv",
"activitynet/videos/v_y7i-jRmrwnI.mp4",
"activitynet/videos/v_UFVeN-ThOwU.mp4",
"activitynet/videos/v_69DNcmkoapw.mp4",
"activitynet/videos/v_QryL-hVKAOA.mp4"
] | 5
|
<answer>3</answer>
|
[
3
] | false
| 0
| 18,223.85288
| 223
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"The little girl is putting the spoon on the other side of the sink. The little girl is brushing the big wooden spoon and put it under the running water and put the wooden spoon on the other side of the sink. The girl picked up the bottle, rub the sponge on it, then rinse it with water, and put it on the strainer. The girl picked up the plate and rub the sponge on plate.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_u8ykXBc2Efs.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_y7i-jRmrwnI.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_UFVeN-ThOwU.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_69DNcmkoapw.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_QryL-hVKAOA.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A woman is at a park with her toddler. She pushes the baby back and forth on a swing. The baby laughs and smiles as she swings."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_iwMXYbYyJy4.mp4",
"activitynet/videos/v_uPqp30C6MDE.mp4",
"activitynet/videos/v_H3PWbSF9ax4.mp4",
"activitynet/videos/v_5I0K3y27EUM.mp4",
"activitynet/videos/v_ZPVrC5185NM.mp4"
] | 5
|
<answer>1</answer>
|
[
1
] | false
| 0
| 18,271.979355
| 225
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A woman is at a park with her toddler. She pushes the baby back and forth on a swing. The baby laughs and smiles as she swings.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_iwMXYbYyJy4.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_uPqp30C6MDE.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_H3PWbSF9ax4.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_5I0K3y27EUM.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_ZPVrC5185NM.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "These people are outside talking and walking along the sidewalk. There are also people running in their short outfits. There's also people watching them as they run in the middle of the street."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_gXp3KSWhf1g.mp4",
"activitynet/videos/v_outMi06JZss.mp4",
"activitynet/videos/v_kUQ4bTeoG-Y.mp4",
"activitynet/videos/v_VlLq4bAHCXI.mp4",
"activitynet/videos/v_duZnMXDWkGw.mp4"
] | 4
|
<answer>3</answer>
|
[
3
] | false
| 0
| 22,123.557007
| 224
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"These people are outside talking and walking along the sidewalk. There are also people running in their short outfits. There's also people watching them as they run in the middle of the street.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_gXp3KSWhf1g.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_outMi06JZss.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_kUQ4bTeoG-Y.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_VlLq4bAHCXI.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_duZnMXDWkGw.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A bartender explains and shows how to prepare exotic alcoholic drinks in glasses using alcohol and juice. Then, the bartender takes a cup and prepares a cocktail in a shaker, then he pours it in the cup. After, the bartender pours liquor in a glass and puts straws and pieces of pineapple in the glass."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_ePaFTey15ho.mp4",
"activitynet/videos/v_9nndNUHadcg.mp4",
"activitynet/videos/v_PyPu-6wATfw.mp4",
"activitynet/videos/v_XlEmG7nM0jw.mp4",
"activitynet/videos/v_pZxteNqdweM.mkv"
] | 3
|
<answer>1</answer>
|
[
1
] | false
| 0
| 15,381.000327
| 226
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A bartender explains and shows how to prepare exotic alcoholic drinks in glasses using alcohol and juice. Then, the bartender takes a cup and prepares a cocktail in a shaker, then he pours it in the cup. After, the bartender pours liquor in a glass and puts straws and pieces of pineapple in the glass.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_ePaFTey15ho.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_9nndNUHadcg.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_PyPu-6wATfw.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_XlEmG7nM0jw.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_pZxteNqdweM.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A man spins a blindfold woman holding a stick who is on front a piñata. A man takes picture to the blindfold woman. The girl try to hit the piñata that is moving around without success. Unexpectedly, the body of the piñata falls while the head is still hung. After, the girl hit the head of the piñata, then she takes off the cloth on his eyes and hit the piñata on the floor."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_3K_8CdJS9lE.mp4",
"activitynet/videos/v_EDkYPikPWW8.mp4",
"activitynet/videos/v_iiqaJGokpEw.mp4",
"activitynet/videos/v_JHHHuKeA-WQ.mp4",
"activitynet/videos/v_SvIUXZqy8Hs.mp4"
] | 4
|
<answer>4</answer>
|
[
4
] | true
| 1
| 14,366.261351
| 228
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A man spins a blindfold woman holding a stick who is on front a piñata. A man takes picture to the blindfold woman. The girl try to hit the piñata that is moving around without success. Unexpectedly, the body of the piñata falls while the head is still hung. After, the girl hit the head of the piñata, then she takes off the cloth on his eyes and hit the piñata on the floor.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_3K_8CdJS9lE.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_EDkYPikPWW8.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_iiqaJGokpEw.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_JHHHuKeA-WQ.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_SvIUXZqy8Hs.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A man is sitting down in a room with a grey back drop and begins talking about the keys on the saxophone he is playing. The man begins to play and the camera continues to zoom in on his fingers that are playing the bottom keys."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_fKDl_CnA8nY.mp4",
"activitynet/videos/v_t9j3GNVm8jw.mp4",
"activitynet/videos/v_oRtMsdNQ1LE.mkv",
"activitynet/videos/v_HrKO4BfXVbk.mp4",
"activitynet/videos/v_9Hxcuf80TK0.mp4"
] | 4
|
<answer>5</answer>
|
[
5
] | false
| 0
| 18,217.333771
| 227
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A man is sitting down in a room with a grey back drop and begins talking about the keys on the saxophone he is playing. The man begins to play and the camera continues to zoom in on his fingers that are playing the bottom keys.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_fKDl_CnA8nY.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_t9j3GNVm8jw.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_oRtMsdNQ1LE.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_HrKO4BfXVbk.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_9Hxcuf80TK0.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A man is seen speaking to the camera followed by shots of a person riding a kayak as well as sitting on a dock and climbing into a kayak. He speaks to the camera more, pulls the kayak out of the water, and walks away with it on his head."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_okvQJRTfGHk.mp4",
"activitynet/videos/v_7ET-TtUVU7s.mp4",
"activitynet/videos/v_uLeJBFypCHE.mkv",
"activitynet/videos/v_Ff8QLpH5T1c.mkv",
"activitynet/videos/v_vD9oh7NZ2PA.mp4"
] | 5
|
<answer>4</answer>
|
[
4
] | false
| 0
| 19,226.252968
| 229
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A man is seen speaking to the camera followed by shots of a person riding a kayak as well as sitting on a dock and climbing into a kayak. He speaks to the camera more, pulls the kayak out of the water, and walks away with it on his head.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_okvQJRTfGHk.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_7ET-TtUVU7s.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_uLeJBFypCHE.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_Ff8QLpH5T1c.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_vD9oh7NZ2PA.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "Several pictures are shown of kids decorating pumpkins that lead into video of a small boy drawing on a pumpkin and a cat sniffing the object. A man then helps the boy cut out the lines of the pumpkin while he looks excited and ends by showing the inside of the pumpkin as well as in the dark."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_9svdYGBSMvM.mp4",
"activitynet/videos/v_PSh-caJvSHU.mp4",
"activitynet/videos/v_YcjLd_XBK5Y.mp4",
"activitynet/videos/v_9IvKkq9k81o.mp4",
"activitynet/videos/v_ZVIi4lPU6h0.mp4"
] | 2
|
<answer>2</answer>
|
[
2
] | true
| 1
| 17,457.525888
| 231
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"Several pictures are shown of kids decorating pumpkins that lead into video of a small boy drawing on a pumpkin and a cat sniffing the object. A man then helps the boy cut out the lines of the pumpkin while he looks excited and ends by showing the inside of the pumpkin as well as in the dark.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_9svdYGBSMvM.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_PSh-caJvSHU.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_YcjLd_XBK5Y.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_9IvKkq9k81o.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_ZVIi4lPU6h0.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "An Athlete competes long jump, first they run fast, skip and jump high in the sand box. A gymnast runs but he do not jump, only walk in the sand box. An athlete walks around waiting for an announce. Then, the athlete jumps happy and raise his hands in sign of victory, he holds a flag and runs while greet people. Cameramen film the the gymnast that win the competition."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_lCIJJgxTs2U.mp4",
"activitynet/videos/v_4o8MaHTb7E4.mp4",
"activitynet/videos/v_ZOQSDsJYXIA.mp4",
"activitynet/videos/v_Dh3bLRYJkiY.mp4",
"activitynet/videos/v_Sx3NHkPp3Jo.mp4"
] | 2
|
<answer>4</answer>
|
[
4
] | false
| 0
| 16,166.878362
| 232
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"An Athlete competes long jump, first they run fast, skip and jump high in the sand box. A gymnast runs but he do not jump, only walk in the sand box. An athlete walks around waiting for an announce. Then, the athlete jumps happy and raise his hands in sign of victory, he holds a flag and runs while greet people. Cameramen film the the gymnast that win the competition.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_lCIJJgxTs2U.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_4o8MaHTb7E4.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_ZOQSDsJYXIA.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_Dh3bLRYJkiY.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_Sx3NHkPp3Jo.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A man is standing in a kitchen taking with a white box in front of him with a silver doily in front of him. He then takes out a tablet that has a picture of Pikachu on it and begins decorating the cake with its face. The cake is finally finished,and the man begins talking and picks up the cake to show the final product."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_rzIaKwWJDZI.mp4",
"activitynet/videos/v_pznmOdbp7E0.mp4",
"activitynet/videos/v_qx1FNJxiUuE.mp4",
"activitynet/videos/v_u5ri43qbi1A.mp4",
"activitynet/videos/v_baSx0q9LKg0.mp4"
] | 4
|
<answer>4</answer>
|
[
4
] | true
| 1
| 26,585.508743
| 230
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A man is standing in a kitchen taking with a white box in front of him with a silver doily in front of him. He then takes out a tablet that has a picture of Pikachu on it and begins decorating the cake with its face. The cake is finally finished,and the man begins talking and picks up the cake to show the final product.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_rzIaKwWJDZI.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_pznmOdbp7E0.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_qx1FNJxiUuE.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_u5ri43qbi1A.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_baSx0q9LKg0.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A woman is on a field with an archery set. She shoots an arrow at a target while other girls watch and wait their turn. Announcers talk as she gets closer to the bullseye. She hits very close to the center, and a fellow player gives her a thumbs up."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_mUyMYnGXKgk.mp4",
"activitynet/videos/v__Wag6CT_0j8.mp4",
"activitynet/videos/v_5rw6n16ILgY.mp4",
"activitynet/videos/v_vTbeVoT1Gsg.mkv",
"activitynet/videos/v_esNQZCjMZaM.mp4"
] | 5
|
<answer>5</answer>
|
[
5
] | true
| 1
| 21,507.777684
| 233
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A woman is on a field with an archery set. She shoots an arrow at a target while other girls watch and wait their turn. Announcers talk as she gets closer to the bullseye. She hits very close to the center, and a fellow player gives her a thumbs up.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_mUyMYnGXKgk.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v__Wag6CT_0j8.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_5rw6n16ILgY.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_vTbeVoT1Gsg.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_esNQZCjMZaM.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "Several people walk onto a field where they are preparing for a game of tug of war. They line up near the rope and begin the competition. They are tugging back and for until one side eventually wins the game."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_wKThOOUV6lY.mp4",
"activitynet/videos/v_mFWRIp164r4.mp4",
"activitynet/videos/v_iMB_mb11KWM.mp4",
"activitynet/videos/v_XELYXH6fqeA.mp4",
"activitynet/videos/v_8r167TmBebg.mp4"
] | 5
|
<answer>1</answer>
|
[
1
] | false
| 0
| 14,464.149209
| 235
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"Several people walk onto a field where they are preparing for a game of tug of war. They line up near the rope and begin the competition. They are tugging back and for until one side eventually wins the game.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_wKThOOUV6lY.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_mFWRIp164r4.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_iMB_mb11KWM.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_XELYXH6fqeA.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_8r167TmBebg.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A woman is seen hosting a news segment with several bottles of sun screen laid out in front of her. Various shots shots are shown of people on the beach putting sun screen on and a skin doctor speaking to the camera. More people are shown on the news segment with the host wearing sun screen and pointing to all the bottles on her table."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_ABQYqpWF1LA.mp4",
"activitynet/videos/v_7S15OsGinjw.mp4",
"activitynet/videos/v_OFKGyZxazQk.mp4",
"activitynet/videos/v_ChPzol03Hqs.mp4",
"activitynet/videos/v_Zzj03Cew2vk.mp4"
] | 3
|
<answer>2</answer>
|
[
2
] | false
| 0
| 20,932.055774
| 234
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A woman is seen hosting a news segment with several bottles of sun screen laid out in front of her. Various shots shots are shown of people on the beach putting sun screen on and a skin doctor speaking to the camera. More people are shown on the news segment with the host wearing sun screen and pointing to all the bottles on her table.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_ABQYqpWF1LA.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_7S15OsGinjw.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_OFKGyZxazQk.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_ChPzol03Hqs.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_Zzj03Cew2vk.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A close up of a weight is shown followed by several clips of people working out intensely. More clips are shown of people working out and looking into the camera while sweating."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_lOtplLrtapE.mp4",
"activitynet/videos/v_ReKUs0km4X8.mp4",
"activitynet/videos/v_cyznGwlE9hM.mp4",
"activitynet/videos/v_L36MIRUpcrI.mp4",
"activitynet/videos/v_RAw8sshR51c.mp4"
] | 4
|
<answer>4</answer>
|
[
4
] | true
| 1
| 19,363.690649
| 236
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A close up of a weight is shown followed by several clips of people working out intensely. More clips are shown of people working out and looking into the camera while sweating.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_lOtplLrtapE.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_ReKUs0km4X8.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_cyznGwlE9hM.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_L36MIRUpcrI.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_RAw8sshR51c.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A man is seen speaking to the camera and walks inbetween a set of uneven bars. The man then lifts himself up above the bars and then jumps back down."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_gXvRxyT5rWw.mp4",
"activitynet/videos/v_DqsaFxxfONY.mp4",
"activitynet/videos/v_TtAEG3yXDnI.mp4",
"activitynet/videos/v_BJ9r8_JnG0k.mp4",
"activitynet/videos/v_xVPTVGpOkGE.mp4"
] | 5
|
<answer>1</answer>
|
[
1
] | false
| 0
| 22,940.195082
| 237
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A man is seen speaking to the camera and walks inbetween a set of uneven bars. The man then lifts himself up above the bars and then jumps back down.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_gXvRxyT5rWw.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_DqsaFxxfONY.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_TtAEG3yXDnI.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_BJ9r8_JnG0k.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_xVPTVGpOkGE.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A man is standing in a room talking. He lays down on his back and does crunches. He stands back up and continues talking."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_qeyFjCAA_dg.mp4",
"activitynet/videos/v_BgAiDS4fF_I.mkv",
"activitynet/videos/v_JOBSEatasv4.mp4",
"activitynet/videos/v_cBAlXvu38dg.mp4",
"activitynet/videos/v_QN6YFgrx_Ig.mp4"
] | 2
|
<answer>2</answer>
|
[
2
] | true
| 1
| 24,338.260821
| 238
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A man is standing in a room talking. He lays down on his back and does crunches. He stands back up and continues talking.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_qeyFjCAA_dg.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_BgAiDS4fF_I.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_JOBSEatasv4.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_cBAlXvu38dg.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_QN6YFgrx_Ig.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A man in a black shirt stands by a glass door. He then begins to rub the frame with a blue rag. As he does this he explains his methods. Using the same rag he moves to the other pane to clean the frame. Finally, he ends by scrubbing the bottom right corner of the door."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_LTPrtyWIcA4.mp4",
"activitynet/videos/v_LV0nevBELso.mp4",
"activitynet/videos/v_jNJg1TYq3c8.mp4",
"activitynet/videos/v_Dk3DiAp2yAU.mp4",
"activitynet/videos/v_9VGbtQrlcN4.mp4"
] | 4
|
<answer>2</answer>
|
[
2
] | false
| 0
| 18,532.81542
| 239
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A man in a black shirt stands by a glass door. He then begins to rub the frame with a blue rag. As he does this he explains his methods. Using the same rag he moves to the other pane to clean the frame. Finally, he ends by scrubbing the bottom right corner of the door.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_LTPrtyWIcA4.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_LV0nevBELso.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_jNJg1TYq3c8.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_Dk3DiAp2yAU.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_9VGbtQrlcN4.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A teenage boy is sitting at a table with his friends and begins sticking his face in his food then they all begin to laugh about it and he gets up. After,all of the boys at the table do the same thing and begin smashing their face and rubbing ice cream around their face."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_LCyLWiw7n5Q.mp4",
"activitynet/videos/v_h4Cf5u1j0TU.mp4",
"activitynet/videos/v_PLnfT1PoVHw.mp4",
"activitynet/videos/v_O5vpeIfQxLQ.mp4",
"activitynet/videos/v_u7OvguFW-Hs.mp4"
] | 1
|
<answer>5</answer>
|
[
5
] | false
| 0
| 13,246.219813
| 240
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A teenage boy is sitting at a table with his friends and begins sticking his face in his food then they all begin to laugh about it and he gets up. After,all of the boys at the table do the same thing and begin smashing their face and rubbing ice cream around their face.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_LCyLWiw7n5Q.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_h4Cf5u1j0TU.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_PLnfT1PoVHw.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_O5vpeIfQxLQ.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_u7OvguFW-Hs.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "The top of a woman's head is seen. She is shown getting her hair braided by another woman. The woman tucks the braid in, showing how she keeps it clipped with bobby pins."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_TIAAUayALPI.mp4",
"activitynet/videos/v_ZGK-w7-bkNw.mp4",
"activitynet/videos/v_470dhR3Yrjs.mp4",
"activitynet/videos/v_Iiwz1JtC7rk.mp4",
"activitynet/videos/v_f1kY1-9XR1k.mp4"
] | 4
|
<answer>4</answer>
|
[
4
] | true
| 1
| 17,655.715789
| 241
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"The top of a woman's head is seen. She is shown getting her hair braided by another woman. The woman tucks the braid in, showing how she keeps it clipped with bobby pins.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_TIAAUayALPI.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_ZGK-w7-bkNw.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_470dhR3Yrjs.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_Iiwz1JtC7rk.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_f1kY1-9XR1k.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "People are standing behind large drums. A man begins playing the drums. He turns next to him and shakes the hand of a man in a suit."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_1AiQt87brik.mp4",
"activitynet/videos/v_8RntjHIwMNo.mp4",
"activitynet/videos/v_pf49xhMRrgQ.mp4",
"activitynet/videos/v_tN1_lOJlUlc.mkv",
"activitynet/videos/v_2uBPhFis_4Y.mp4"
] | 2
|
<answer>4</answer>
|
[
4
] | false
| 0
| 16,717.262025
| 242
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"People are standing behind large drums. A man begins playing the drums. He turns next to him and shakes the hand of a man in a suit.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_1AiQt87brik.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_8RntjHIwMNo.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_pf49xhMRrgQ.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_tN1_lOJlUlc.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_2uBPhFis_4Y.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "a woman is on a high beam. She flips forward and backward multiple times as her trainer and other girls watch. She performs elegantly, jumping into the air and performing flips before dismounting."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_H1bmoIihWwo.mp4",
"activitynet/videos/v_bmWICdhvyJw.mp4",
"activitynet/videos/v_OS-h1xzAZno.mp4",
"activitynet/videos/v_PZ1FVhgTRWU.mp4",
"activitynet/videos/v_5QbiJmDyoM0.mp4"
] | 2
|
<answer>5</answer>
|
[
5
] | false
| 0
| 18,117.214991
| 243
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"a woman is on a high beam. She flips forward and backward multiple times as her trainer and other girls watch. She performs elegantly, jumping into the air and performing flips before dismounting.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_H1bmoIihWwo.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_bmWICdhvyJw.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_OS-h1xzAZno.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_PZ1FVhgTRWU.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_5QbiJmDyoM0.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A man is seen speaking to the camera while holding a tennis racket. The man is then shown in several shots hitting a ball around a closed in area with another person. The man continues to hit the ball around the room with another person while speaking to the camera."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_h1t5QZjERms.mp4",
"activitynet/videos/v_PCSlAOYPMOs.mkv",
"activitynet/videos/v_aotVhoXjqS0.mp4",
"activitynet/videos/v_akwJwcvfjLA.mp4",
"activitynet/videos/v_3TxZTZEEg44.mp4"
] | 4
|
<answer>5</answer>
|
[
5
] | false
| 0
| 20,795.964139
| 244
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A man is seen speaking to the camera while holding a tennis racket. The man is then shown in several shots hitting a ball around a closed in area with another person. The man continues to hit the ball around the room with another person while speaking to the camera.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_h1t5QZjERms.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_PCSlAOYPMOs.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_aotVhoXjqS0.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_akwJwcvfjLA.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_3TxZTZEEg44.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A man is pouring water into a white bucket. He sprays water on the car. He dips a towel into the bucket and starts washing the car."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_z8VqGGu5vPc.mp4",
"activitynet/videos/v_E-M2Cq0RNTs.mp4",
"activitynet/videos/v_g0upuaWM74M.mp4",
"activitynet/videos/v_mio5dnRbo4w.mkv",
"activitynet/videos/v_z08g5S7J-CY.mp4"
] | 2
|
<answer>4</answer>
|
[
4
] | false
| 0
| 17,377.187518
| 246
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A man is pouring water into a white bucket. He sprays water on the car. He dips a towel into the bucket and starts washing the car.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_z8VqGGu5vPc.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_E-M2Cq0RNTs.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_g0upuaWM74M.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_mio5dnRbo4w.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_z08g5S7J-CY.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A man is sitting outside of a wall playing what looks like a miniature piano attached to a bag with his case out for money as a crowd of people surround him to watch. As he plays,a tall man dressed in all blacks then jumps behind him and starts to clap to get some attention."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_vCAGiXqYXBk.mp4",
"activitynet/videos/v_5GFpN0YZEog.mp4",
"activitynet/videos/v_pf49xhMRrgQ.mp4",
"activitynet/videos/v_prqwtY9cn6s.mp4",
"activitynet/videos/v_yzN9jN3qncA.mp4"
] | 1
|
<answer>1</answer>
|
[
1
] | true
| 1
| 23,908.009668
| 245
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A man is sitting outside of a wall playing what looks like a miniature piano attached to a bag with his case out for money as a crowd of people surround him to watch. As he plays,a tall man dressed in all blacks then jumps behind him and starts to clap to get some attention.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_vCAGiXqYXBk.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_5GFpN0YZEog.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_pf49xhMRrgQ.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_prqwtY9cn6s.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_yzN9jN3qncA.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A woman is swinging an ax at a tree stump. She drops the ax on the ground. She picks it up and continues swinging."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_7QxUtHqQdbY.mp4",
"activitynet/videos/v_Pt5jMqQXTZ8.mp4",
"activitynet/videos/v_QnQ2D-tJ9pM.mp4",
"activitynet/videos/v_PKYg6_rs3LQ.mp4",
"activitynet/videos/v_Lomlff9wClo.mp4"
] | 5
|
<answer>5</answer>
|
[
5
] | true
| 1
| 9,461.303868
| 247
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A woman is swinging an ax at a tree stump. She drops the ax on the ground. She picks it up and continues swinging.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_7QxUtHqQdbY.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_Pt5jMqQXTZ8.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_QnQ2D-tJ9pM.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_PKYg6_rs3LQ.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_Lomlff9wClo.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A woman is seen speaking to the camera and leads to her holding up mouth wash and then swishing it around her mouth. She continues swishing it around while closing the cap then spits it out and smiles to the camera."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_Uj1QtIM8500.mp4",
"activitynet/videos/v_J4hnBPgwDlw.mp4",
"activitynet/videos/v_uTQyPHg8r0M.mp4",
"activitynet/videos/v_ffUtqOyJ7fM.mp4",
"activitynet/videos/v_EFGtb9IDQao.mp4"
] | 2
|
<answer>2</answer>
|
[
2
] | true
| 1
| 11,184.274816
| 248
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A woman is seen speaking to the camera and leads to her holding up mouth wash and then swishing it around her mouth. She continues swishing it around while closing the cap then spits it out and smiles to the camera.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_Uj1QtIM8500.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_J4hnBPgwDlw.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_uTQyPHg8r0M.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_ffUtqOyJ7fM.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_EFGtb9IDQao.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A chef is seen speaking to the camera behind a kitchen counter and leads into him cutting up lettuce and pouring ingredients out into a bowl. The man presents more food items on a plate as well as a bowl sitting next to it."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_i4yQ54eWfy4.mp4",
"activitynet/videos/v_I2Y-5EEXAE0.mp4",
"activitynet/videos/v_upoS4Jct7kE.mp4",
"activitynet/videos/v_ux_qqONPSrc.mp4",
"activitynet/videos/v_tVC_5_SgseY.mp4"
] | 1
|
<answer>5</answer>
|
[
5
] | false
| 0
| 14,774.229839
| 249
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A chef is seen speaking to the camera behind a kitchen counter and leads into him cutting up lettuce and pouring ingredients out into a bowl. The man presents more food items on a plate as well as a bowl sitting next to it.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_i4yQ54eWfy4.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_I2Y-5EEXAE0.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_upoS4Jct7kE.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_ux_qqONPSrc.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_tVC_5_SgseY.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A man opens an icy car door and snow falls off. The man opens and closes the door a couple of times. The man talks to the camera then scrapes the ice off the drivers side window. The man turns to the camera for a moment. The man then cleans the side mirror."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_eM2miz5uf8Q.mp4",
"activitynet/videos/v_TEh6gfRUFZQ.mp4",
"activitynet/videos/v_YCqbvmEG-Uw.mp4",
"activitynet/videos/v_SSJjjggYBxc.mp4",
"activitynet/videos/v_t1urvYx1X_w.mp4"
] | 1
|
<answer>2</answer>
|
[
2
] | false
| 0
| 12,017.408206
| 251
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A man opens an icy car door and snow falls off. The man opens and closes the door a couple of times. The man talks to the camera then scrapes the ice off the drivers side window. The man turns to the camera for a moment. The man then cleans the side mirror.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_eM2miz5uf8Q.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_TEh6gfRUFZQ.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_YCqbvmEG-Uw.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_SSJjjggYBxc.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_t1urvYx1X_w.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "Two men are dancing on a basketball court. A man does a back flip on a basketball court. The two men do the splits on the court."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_8VPjByN_v9w.mp4",
"activitynet/videos/v_7-_Nur_xiV4.mp4",
"activitynet/videos/v_NzxZdC-63LE.mp4",
"activitynet/videos/v_bY-4XBIGiwI.mp4",
"activitynet/videos/v_-VexUX6OJBM.mp4"
] | 1
|
<answer>1</answer>
|
[
1
] | true
| 1
| 19,974.757282
| 250
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"Two men are dancing on a basketball court. A man does a back flip on a basketball court. The two men do the splits on the court.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_8VPjByN_v9w.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_7-_Nur_xiV4.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_NzxZdC-63LE.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_bY-4XBIGiwI.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_-VexUX6OJBM.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A young woman is seen standing behind a netting area with a swing in her hand and moving in slow motion. She then throws the object around while she spins and finally throw it off into the distance while looking away."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_MiOJxYa5Nt4.mp4",
"activitynet/videos/v_Q0U51Hqn21w.mp4",
"activitynet/videos/v_Y5uVICaJU-0.mp4",
"activitynet/videos/v_mpC_UTM1tWQ.mp4",
"activitynet/videos/v_x8cuLOUppmU.mp4"
] | 5
|
<answer>5</answer>
|
[
5
] | true
| 1
| 11,358.503335
| 252
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A young woman is seen standing behind a netting area with a swing in her hand and moving in slow motion. She then throws the object around while she spins and finally throw it off into the distance while looking away.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_MiOJxYa5Nt4.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_Q0U51Hqn21w.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_Y5uVICaJU-0.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_mpC_UTM1tWQ.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_x8cuLOUppmU.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A girl is shown playing volleyball on a team in an indoor gym. The ball is lobbed back and forth over the net multiple times."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_sCj-ME5RkLY.mp4",
"activitynet/videos/v_cICxG-28hK0.mp4",
"activitynet/videos/v_-A6e83tl4Y8.mp4",
"activitynet/videos/v_IEPoIqIrprg.mp4",
"activitynet/videos/v_RhOV_K2XzZA.mp4"
] | 1
|
<answer>3</answer>
|
[
3
] | false
| 0
| 12,109.382957
| 253
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A girl is shown playing volleyball on a team in an indoor gym. The ball is lobbed back and forth over the net multiple times.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_sCj-ME5RkLY.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_cICxG-28hK0.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_-A6e83tl4Y8.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_IEPoIqIrprg.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_RhOV_K2XzZA.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "People spin around and throw balls onto a field. The crowd watches them throw the balls. A man is laying on his back lifting a large weight over his body."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_HYAlS44yzdo.mp4",
"activitynet/videos/v_LAZHNzFbDNY.mp4",
"activitynet/videos/v_cyznGwlE9hM.mp4",
"activitynet/videos/v_62Dwj4l7_qs.mp4",
"activitynet/videos/v_KbbEbeCJTJg.mp4"
] | 3
|
<answer>2</answer>
|
[
2
] | false
| 0
| 15,972.34551
| 254
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"People spin around and throw balls onto a field. The crowd watches them throw the balls. A man is laying on his back lifting a large weight over his body.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_HYAlS44yzdo.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_LAZHNzFbDNY.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_cyznGwlE9hM.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_62Dwj4l7_qs.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_KbbEbeCJTJg.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A man pushes a toddler in a swing. A boy jumps knelled behind the toddler. Then, the boy walks and pass on side the toddler."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_m5PO3T2uGzs.mp4",
"activitynet/videos/v_kU2FVf0ldx4.mp4",
"activitynet/videos/v_9KPRS9y8Fvo.mp4",
"activitynet/videos/v_u9ec3Exc5mI.mp4",
"activitynet/videos/v_lLHAzwAs_9I.mp4"
] | 1
|
<answer>5</answer>
|
[
5
] | false
| 0
| 19,256.39927
| 255
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A man pushes a toddler in a swing. A boy jumps knelled behind the toddler. Then, the boy walks and pass on side the toddler.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_m5PO3T2uGzs.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_kU2FVf0ldx4.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_9KPRS9y8Fvo.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_u9ec3Exc5mI.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_lLHAzwAs_9I.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "Several girls are in a classroom dancing and doing ballet. The instructor then comes to talk briefly before continuing on coaching the girls. After,the exercises continue and the girls do leaps and jumps in the room before the outside of the dance studio is shown."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_iWSKl7vOd2s.mp4",
"activitynet/videos/v_zmmiX3_TJ84.mp4",
"activitynet/videos/v_VDYSVR0HbpM.mp4",
"activitynet/videos/v_bBtzyRzk0UM.mp4",
"activitynet/videos/v_bmIWsU8sNlw.mkv"
] | 1
|
<answer>4</answer>
|
[
4
] | false
| 0
| 17,758.599051
| 256
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"Several girls are in a classroom dancing and doing ballet. The instructor then comes to talk briefly before continuing on coaching the girls. After,the exercises continue and the girls do leaps and jumps in the room before the outside of the dance studio is shown.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_iWSKl7vOd2s.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_zmmiX3_TJ84.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_VDYSVR0HbpM.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_bBtzyRzk0UM.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_bmIWsU8sNlw.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "Three kids are at the table, the little girl at the center is mixing the bowl while dancing. The kids are licking their fingers and spatula. The girls are adding ingredients in the bowl and then mixing the ingredients at the same time. The kids wiped the platter, the adult put the batter in the oven and then took it out then the kids added decorations on the cake."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_vc820BteGzY.mp4",
"activitynet/videos/v_8zq6C0SRyDQ.mp4",
"activitynet/videos/v_F3jJVS3NHf8.mp4",
"activitynet/videos/v_Hhc10CrukfA.mp4",
"activitynet/videos/v_Pv6oIFroaCQ.mp4"
] | 1
|
<answer>2</answer>
|
[
2
] | false
| 0
| 19,798.560209
| 257
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"Three kids are at the table, the little girl at the center is mixing the bowl while dancing. The kids are licking their fingers and spatula. The girls are adding ingredients in the bowl and then mixing the ingredients at the same time. The kids wiped the platter, the adult put the batter in the oven and then took it out then the kids added decorations on the cake.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_vc820BteGzY.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_8zq6C0SRyDQ.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_F3jJVS3NHf8.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_Hhc10CrukfA.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_Pv6oIFroaCQ.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A man applies plaster to the ceiling with tools while two others work at other tasks in the background. A closeup of the ceiling section is shown while the man gets more plaster. The man continues to apply plaster to the ceiling. The talks to the camera as he gets more plaster. The man resumes plastering the ceiling. One of the other men talk to the man plastering the ceiling. The first man stops plastering and responds to the second man. The second man continues to talk to the first man while using a pipe as a penis. The first man stops plastering and talks to the second man some more. The first man laughs at the camera while holding the bench he had been standing on."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_HaGLPOqibaM.mp4",
"activitynet/videos/v_nEmuDmbOp1E.mp4",
"activitynet/videos/v_Ny49eEt1OJg.mp4",
"activitynet/videos/v_rczR9C00KOM.mp4",
"activitynet/videos/v_cXY-ONmtylc.mp4"
] | 4
|
<answer>1</answer>
|
[
1
] | false
| 0
| 19,221.033079
| 258
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A man applies plaster to the ceiling with tools while two others work at other tasks in the background. A closeup of the ceiling section is shown while the man gets more plaster. The man continues to apply plaster to the ceiling. The talks to the camera as he gets more plaster. The man resumes plastering the ceiling. One of the other men talk to the man plastering the ceiling. The first man stops plastering and responds to the second man. The second man continues to talk to the first man while using a pipe as a penis. The first man stops plastering and talks to the second man some more. The first man laughs at the camera while holding the bench he had been standing on.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_HaGLPOqibaM.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_nEmuDmbOp1E.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_Ny49eEt1OJg.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_rczR9C00KOM.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_cXY-ONmtylc.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "People are playing lacrosse on a field. A girl in a maroon shirt is hitting the ball with her stick. Another girl in a yellow shirt is standing on the field."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_nWpFumm3Z5g.mp4",
"activitynet/videos/v__Wag6CT_0j8.mp4",
"activitynet/videos/v_er6fi7nYsuw.mp4",
"activitynet/videos/v_6hu3V1PS4vM.mp4",
"activitynet/videos/v_WKXIl7wvlk0.mp4"
] | 3
|
<answer>5</answer>
|
[
5
] | false
| 0
| 18,187.280257
| 259
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"People are playing lacrosse on a field. A girl in a maroon shirt is hitting the ball with her stick. Another girl in a yellow shirt is standing on the field.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_nWpFumm3Z5g.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v__Wag6CT_0j8.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_er6fi7nYsuw.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_6hu3V1PS4vM.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_WKXIl7wvlk0.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "An African American toddler's right hand is placed on a white rag while another young girl pants her nails with burgundy polish. The toddler with the brush continues to dip the brush in the pain and paints her friend's nail while laughing. After, the girl looks to the camera and begins smiling."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_DvDfifKGXXg.mp4",
"activitynet/videos/v_ZJk05q3y5iM.mp4",
"activitynet/videos/v_xKePBw5XZHs.mp4",
"activitynet/videos/v_JMlNfZlOyX8.mp4",
"activitynet/videos/v_fZQS02Ypca4.mp4"
] | 2
|
<answer>2</answer>
|
[
2
] | true
| 1
| 11,865.142429
| 260
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"An African American toddler's right hand is placed on a white rag while another young girl pants her nails with burgundy polish. The toddler with the brush continues to dip the brush in the pain and paints her friend's nail while laughing. After, the girl looks to the camera and begins smiling.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_DvDfifKGXXg.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_ZJk05q3y5iM.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_xKePBw5XZHs.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_JMlNfZlOyX8.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_fZQS02Ypca4.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "men are fighting bulls on a field. people are sitting around the court watching the bullfighter. two bulls are running in a green court."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_MHhMO3yhcfU.mkv",
"activitynet/videos/v_Gg32cIypcdc.mkv",
"activitynet/videos/v_NzrOOXRyDPM.mp4",
"activitynet/videos/v_Qnr73D2zIjU.mp4",
"activitynet/videos/v_xhBvsWa0PCs.mp4"
] | 4
|
<answer>1</answer>
|
[
1
] | false
| 0
| 13,653.943762
| 262
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"men are fighting bulls on a field. people are sitting around the court watching the bullfighter. two bulls are running in a green court.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_MHhMO3yhcfU.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_Gg32cIypcdc.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_NzrOOXRyDPM.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_Qnr73D2zIjU.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_xhBvsWa0PCs.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "An athletic man by the name of DAVID DURANTE chalks his hands, walks toward his ramp, raises his right arm, walks onto the ramp, raises both arms, then jumps onto the double ropes while the person behind him pulls the ramp away as soon as he jumps. The man then does his double rope routine that includes a lot of handstands, flips, twists and various other tricks. When the man is done with his routine he does multiple spins and jumps to the side of the ropes, lands on his feet, raises his two arms, claps and turns, raises his two arms again, cheers himself on and walks away while waving and celebrating with high fives and hugs from the people around him."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_cKMGacBQX0E.mp4",
"activitynet/videos/v_pKFBr0pMn7I.mp4",
"activitynet/videos/v_OzXD3WO6jrs.mp4",
"activitynet/videos/v_9gU5be5YCVw.mp4",
"activitynet/videos/v_nEj34gf508E.mp4"
] | 5
|
<answer>5</answer>
|
[
5
] | true
| 1
| 16,775.126972
| 261
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"An athletic man by the name of DAVID DURANTE chalks his hands, walks toward his ramp, raises his right arm, walks onto the ramp, raises both arms, then jumps onto the double ropes while the person behind him pulls the ramp away as soon as he jumps. The man then does his double rope routine that includes a lot of handstands, flips, twists and various other tricks. When the man is done with his routine he does multiple spins and jumps to the side of the ropes, lands on his feet, raises his two arms, claps and turns, raises his two arms again, cheers himself on and walks away while waving and celebrating with high fives and hugs from the people around him.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_cKMGacBQX0E.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_pKFBr0pMn7I.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_OzXD3WO6jrs.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_9gU5be5YCVw.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_nEj34gf508E.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A small child and a large man arm wrestle in front of a crowd. The small child beats the man. The man throws a fit. The audience cheers the small child."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_jQVT4u6NojM.mp4",
"activitynet/videos/v_clSku91LoQQ.mp4",
"activitynet/videos/v_DXhVbxfmrYM.mp4",
"activitynet/videos/v_eMI2x3HFozQ.mp4",
"activitynet/videos/v_jeaaS1NK_d4.mp4"
] | 4
|
<answer>4</answer>
|
[
4
] | true
| 1
| 17,805.941929
| 263
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A small child and a large man arm wrestle in front of a crowd. The small child beats the man. The man throws a fit. The audience cheers the small child.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_jQVT4u6NojM.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_clSku91LoQQ.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_DXhVbxfmrYM.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_eMI2x3HFozQ.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_jeaaS1NK_d4.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "The video leads into clips of two people performing martial arts to one another while a group of people watch and clap. The men continue spinning around one another and ends with a picture and a man walking away."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_6LLDsbc8XMM.mp4",
"activitynet/videos/v_svWiQtzgtOc.mp4",
"activitynet/videos/v_m2hiQ9EOUUI.mp4",
"activitynet/videos/v_ma9R2AjCRZE.mp4",
"activitynet/videos/v_MjljlkQaHh4.mp4"
] | 1
|
<answer>2</answer>
|
[
2
] | false
| 0
| 11,814.851232
| 264
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"The video leads into clips of two people performing martial arts to one another while a group of people watch and clap. The men continue spinning around one another and ends with a picture and a man walking away.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_6LLDsbc8XMM.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_svWiQtzgtOc.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_m2hiQ9EOUUI.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_ma9R2AjCRZE.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_MjljlkQaHh4.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "This video shows viewers how to apply self stick wallpaper. First he applies the wallpaper and explains how to cut it properly. Then he gives other helpful tips of how to put the wallpaper onto the wall and he smoothens it out."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_OsrRpGbIpKA.mp4",
"activitynet/videos/v_vL8s-b4eJiU.mp4",
"activitynet/videos/v_TjLoGNBzNRA.mp4",
"activitynet/videos/v_ep2Kyk8CHT8.mp4",
"activitynet/videos/v_R3MPcPKQYKE.mp4"
] | 2
|
<answer>2</answer>
|
[
2
] | true
| 1
| 14,931.790965
| 265
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"This video shows viewers how to apply self stick wallpaper. First he applies the wallpaper and explains how to cut it properly. Then he gives other helpful tips of how to put the wallpaper onto the wall and he smoothens it out.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_OsrRpGbIpKA.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_vL8s-b4eJiU.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_TjLoGNBzNRA.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_ep2Kyk8CHT8.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_R3MPcPKQYKE.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A man and a woman sail in a boat down the river while moving the oars. People in boats are on more calm waters."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_g17h49EYsJY.mp4",
"activitynet/videos/v_qH3HnhEaeok.mkv",
"activitynet/videos/v_yWEFVfX-JoI.mp4",
"activitynet/videos/v_uevUOX7Wpz4.mkv",
"activitynet/videos/v_RPkH81M6-NE.mp4"
] | 4
|
<answer>4</answer>
|
[
4
] | true
| 1
| 14,700.045366
| 266
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A man and a woman sail in a boat down the river while moving the oars. People in boats are on more calm waters.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_g17h49EYsJY.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_qH3HnhEaeok.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_yWEFVfX-JoI.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_uevUOX7Wpz4.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_RPkH81M6-NE.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A man is seated in a barber chair. He is wearing a cape and looking into a mirror. A woman is styling and cutting his hair."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_-l16smV_uYg.mp4",
"activitynet/videos/v_1VwNfMlb4JU.mp4",
"activitynet/videos/v_WqnnGmL-lmU.mp4",
"activitynet/videos/v_30Yk_1Yc7Vk.mp4",
"activitynet/videos/v_X82bc2v5kcM.mp4"
] | 4
|
<answer>2</answer>
|
[
2
] | false
| 0
| 16,081.086271
| 267
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A man is seated in a barber chair. He is wearing a cape and looking into a mirror. A woman is styling and cutting his hair.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_-l16smV_uYg.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_1VwNfMlb4JU.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_WqnnGmL-lmU.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_30Yk_1Yc7Vk.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_X82bc2v5kcM.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "An intro of what seems to be an Indian man is shown and documenting his everyday life. The man is then shown playing on a small organ with a microphone attached to it. The camera then moves to capture the vacancy of the room and the man is shown playing the flute. After,the intro video is played again and the credits begin to roll."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_ymmBQHiNK24.mp4",
"activitynet/videos/v_wdD-UHM8rTg.mp4",
"activitynet/videos/v_Q7cgJD7-sEM.mp4",
"activitynet/videos/v_mDaZqz7lB0o.mp4",
"activitynet/videos/v_r8AXq1Q5bn0.mp4"
] | 4
|
<answer>3</answer>
|
[
3
] | false
| 0
| 15,087.612874
| 268
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"An intro of what seems to be an Indian man is shown and documenting his everyday life. The man is then shown playing on a small organ with a microphone attached to it. The camera then moves to capture the vacancy of the room and the man is shown playing the flute. After,the intro video is played again and the credits begin to roll.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_ymmBQHiNK24.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_wdD-UHM8rTg.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_Q7cgJD7-sEM.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_mDaZqz7lB0o.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_r8AXq1Q5bn0.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A man is seen speaking to the camera and leads into him punching and kicking a boxing bag. The man continues hitting the bag with his hands and feet while the camera still watches him."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_gVixuVE0-ek.mp4",
"activitynet/videos/v_EIibo7aTpys.mp4",
"activitynet/videos/v_sZRUTtoxY_s.mp4",
"activitynet/videos/v_7fwrkFHTm-Q.mp4",
"activitynet/videos/v_g49F9coR2VU.mp4"
] | 3
|
<answer>1</answer>
|
[
1
] | false
| 0
| 12,794.36452
| 269
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A man is seen speaking to the camera and leads into him punching and kicking a boxing bag. The man continues hitting the bag with his hands and feet while the camera still watches him.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_gVixuVE0-ek.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_EIibo7aTpys.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_sZRUTtoxY_s.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_7fwrkFHTm-Q.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_g49F9coR2VU.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "This woman is twirling her baton and doing cartwheels and such, trying to persuade how baton is a sport. They give plenty of reasons why they consider baton to be a sport. Women and men can do it, they can practice in the streets, they can perform for judges and as duet. One woman even handles her batons while they are on fire, its very impressive."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_GSFyEkGCUVo.mp4",
"activitynet/videos/v_lI6h3H4Zs98.mp4",
"activitynet/videos/v_WRc1Jv1j3nk.mp4",
"activitynet/videos/v_BDQHEemWnSk.mp4",
"activitynet/videos/v_XIMi2oydVB8.mp4"
] | 1
|
<answer>5</answer>
|
[
5
] | false
| 0
| 17,872.091677
| 270
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"This woman is twirling her baton and doing cartwheels and such, trying to persuade how baton is a sport. They give plenty of reasons why they consider baton to be a sport. Women and men can do it, they can practice in the streets, they can perform for judges and as duet. One woman even handles her batons while they are on fire, its very impressive.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_GSFyEkGCUVo.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_lI6h3H4Zs98.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_WRc1Jv1j3nk.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_BDQHEemWnSk.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_XIMi2oydVB8.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A woman is riding on an eliiptical as advertisements for the equipment flash across the screen. She uses the trainer from multiple angels, pushing and pulling while pedaling."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_swbCsf51XVg.mp4",
"activitynet/videos/v_pnFRC2_HPrE.mp4",
"activitynet/videos/v_zpBZ7HMNO34.mp4",
"activitynet/videos/v__MR8G1jwM4o.mp4",
"activitynet/videos/v_15IRaGI4Ml0.mkv"
] | 3
|
<answer>1</answer>
|
[
1
] | false
| 0
| 17,466.151765
| 271
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A woman is riding on an eliiptical as advertisements for the equipment flash across the screen. She uses the trainer from multiple angels, pushing and pulling while pedaling.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_swbCsf51XVg.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_pnFRC2_HPrE.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_zpBZ7HMNO34.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v__MR8G1jwM4o.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_15IRaGI4Ml0.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "The video starts out with several different soccer players showing highlights of incredible goals they've made. The video continues with several more amazing soccer moves being shown with players high fiving and cheering in the end."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_p81NOkb2rww.mp4",
"activitynet/videos/v_RjFoJggnfj4.mp4",
"activitynet/videos/v_Likt_9dbMqE.mp4",
"activitynet/videos/v_j1XZ3FA8EYY.mkv",
"activitynet/videos/v_UHNUmpx0nww.mp4"
] | 4
|
<answer>2</answer>
|
[
2
] | false
| 0
| 20,708.77687
| 272
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"The video starts out with several different soccer players showing highlights of incredible goals they've made. The video continues with several more amazing soccer moves being shown with players high fiving and cheering in the end.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_p81NOkb2rww.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_RjFoJggnfj4.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_Likt_9dbMqE.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_j1XZ3FA8EYY.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_UHNUmpx0nww.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A young man is seen speaking to the camera and begins putting ingredients into a blender. He blends them up together and takes a drink afterwards. He shows off the drink and speaks to the camera."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_5hXH-TorJ6M.mp4",
"activitynet/videos/v_IMto8gJvRek.mp4",
"activitynet/videos/v_6RePzOd3GvQ.mp4",
"activitynet/videos/v_MVzypK0eMKc.mp4",
"activitynet/videos/v_rzm4V_McRhQ.mp4"
] | 4
|
<answer>1</answer>
|
[
1
] | false
| 0
| 16,524.922271
| 273
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A young man is seen speaking to the camera and begins putting ingredients into a blender. He blends them up together and takes a drink afterwards. He shows off the drink and speaks to the camera.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_5hXH-TorJ6M.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_IMto8gJvRek.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_6RePzOd3GvQ.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_MVzypK0eMKc.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_rzm4V_McRhQ.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A young girl climbs a tall diving board and adjusts it. She then dives backwards into the indoor pool. Judges score her performance from the sidelines."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_ccKJg_f1UDo.mp4",
"activitynet/videos/v_ynvCxrj1UNg.mp4",
"activitynet/videos/v_FtbrPGaINt0.mp4",
"activitynet/videos/v_H1bmoIihWwo.mp4",
"activitynet/videos/v_M_WEOecjwLY.mkv"
] | 4
|
<answer>2</answer>
|
[
2
] | false
| 0
| 20,177.407822
| 274
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A young girl climbs a tall diving board and adjusts it. She then dives backwards into the indoor pool. Judges score her performance from the sidelines.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_ccKJg_f1UDo.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_ynvCxrj1UNg.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_FtbrPGaINt0.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_H1bmoIihWwo.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_M_WEOecjwLY.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "Words come on the screen before the video. Action figures are shown on a table. One of the actin figures is knocked down."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_5zT1GWfmVLU.mp4",
"activitynet/videos/v_5n8wY8hwy3Y.mp4",
"activitynet/videos/v_Keuj_3QyLq0.mp4",
"activitynet/videos/v_0ZXc2fEDgg8.mp4",
"activitynet/videos/v_b6VAlwv45q4.mkv"
] | 3
|
<answer>3</answer>
|
[
3
] | true
| 1
| 17,586.221517
| 275
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"Words come on the screen before the video. Action figures are shown on a table. One of the actin figures is knocked down.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_5zT1GWfmVLU.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_5n8wY8hwy3Y.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_Keuj_3QyLq0.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_0ZXc2fEDgg8.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_b6VAlwv45q4.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A person puts their finger in front of the camera and there are people walking around. Then someone runs on the track and jumps towards the end into the sand. Then they stop and people are shown sitting in the grass talking to each other and doing other things."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_huUb8mM5fv4.mp4",
"activitynet/videos/v_Ih8bPM3p0rE.mp4",
"activitynet/videos/v_I4kjOE8HnU0.mp4",
"activitynet/videos/v_IajP-SB2D5c.mp4",
"activitynet/videos/v_GxOjqC_IDX4.mp4"
] | 1
|
<answer>4</answer>
|
[
4
] | false
| 0
| 18,408.465987
| 276
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A person puts their finger in front of the camera and there are people walking around. Then someone runs on the track and jumps towards the end into the sand. Then they stop and people are shown sitting in the grass talking to each other and doing other things.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_huUb8mM5fv4.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_Ih8bPM3p0rE.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_I4kjOE8HnU0.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_IajP-SB2D5c.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_GxOjqC_IDX4.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A young man is standing bent over on a grassy field. The man hits a croquet ball with a mallet and it strikes a wooden stick. The man celebrates by repeatedly raising his hands in victory."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_2q_4I3ae0J4.mp4",
"activitynet/videos/v_gA0m7YUH408.mp4",
"activitynet/videos/v_qrvPTE0kb5U.mp4",
"activitynet/videos/v_axzmwzPQ134.mp4",
"activitynet/videos/v_IQGg87yZZjs.mp4"
] | 2
|
<answer>2</answer>
|
[
2
] | true
| 1
| 18,528.558178
| 277
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A young man is standing bent over on a grassy field. The man hits a croquet ball with a mallet and it strikes a wooden stick. The man celebrates by repeatedly raising his hands in victory.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_2q_4I3ae0J4.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_gA0m7YUH408.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_qrvPTE0kb5U.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_axzmwzPQ134.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_IQGg87yZZjs.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A brown rabbit with a white belly is sitting on the ground and grooming itself while licking its right side. When the rabbit stops licking its right side it briefly faces forward and its left back foot scratches its left ear then he licks his left paw. The rabbit then lifts its right paw and scratches his right ear and then licks that paw. The rabbit then begins to take its front paws and starts rubbing it's face and ears and licking its front paws over and over. Finally the rabbit is done grooming itself and it just sits still."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_NcTZ3wgdNOQ.mp4",
"activitynet/videos/v_tBFX7g605Go.mp4",
"activitynet/videos/v_Feo8xSjY5A8.mp4",
"activitynet/videos/v_K5wPwCFVkhU.mp4",
"activitynet/videos/v_RN2QwhcAsUY.mp4"
] | 2
|
<answer>2</answer>
|
[
2
] | true
| 1
| 14,206.535628
| 278
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A brown rabbit with a white belly is sitting on the ground and grooming itself while licking its right side. When the rabbit stops licking its right side it briefly faces forward and its left back foot scratches its left ear then he licks his left paw. The rabbit then lifts its right paw and scratches his right ear and then licks that paw. The rabbit then begins to take its front paws and starts rubbing it's face and ears and licking its front paws over and over. Finally the rabbit is done grooming itself and it just sits still.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_NcTZ3wgdNOQ.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_tBFX7g605Go.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_Feo8xSjY5A8.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_K5wPwCFVkhU.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_RN2QwhcAsUY.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "Butter is blended in a bowl. Sugar is added to the bowl. Chocolate is added to the mix. A liquid is also added to it."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_drzTgrfN19M.mp4",
"activitynet/videos/v_nKa1e_CpvoY.mp4",
"activitynet/videos/v_v7OluHKECRA.mp4",
"activitynet/videos/v_krNVpENNPCM.mp4",
"activitynet/videos/v_U0d68z5HTwE.mp4"
] | 3
|
<answer>5</answer>
|
[
5
] | false
| 0
| 13,588.553135
| 279
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"Butter is blended in a bowl. Sugar is added to the bowl. Chocolate is added to the mix. A liquid is also added to it.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_drzTgrfN19M.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_nKa1e_CpvoY.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_v7OluHKECRA.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_krNVpENNPCM.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_U0d68z5HTwE.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A man wearing black shirt and gray pants is in the gym carrying a barbel with two 15 lbs. black metal plates, he lifted the barbel then carried up all the way over his head."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_j9QPrMZuegY.mp4",
"activitynet/videos/v_rRkwB9EcEMs.mp4",
"activitynet/videos/v_PY6WgOIZlhw.mp4",
"activitynet/videos/v_70bS0DkAeDo.mp4",
"activitynet/videos/v_kNUpypAppjk.mp4"
] | 5
|
<answer>5</answer>
|
[
5
] | true
| 1
| 17,849.827798
| 280
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A man wearing black shirt and gray pants is in the gym carrying a barbel with two 15 lbs. black metal plates, he lifted the barbel then carried up all the way over his head.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_j9QPrMZuegY.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_rRkwB9EcEMs.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_PY6WgOIZlhw.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_70bS0DkAeDo.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_kNUpypAppjk.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A man is on top of the roof talking about how to work up there using certain tools and putting them under the shingles. He was able to get one out from underneath and then begins to press down on some metal. He gets a hammer and begins to nail it in place and then he cuts off the excess. He lays down a new shingle to replace it over the metal piece he previously nailed in."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_mNM-JUC7ZEA.mp4",
"activitynet/videos/v_eBlYGGmeBY0.mp4",
"activitynet/videos/v_Oc8ACBiwIyE.mp4",
"activitynet/videos/v_LAicExwwM54.mp4",
"activitynet/videos/v_xf9iLflgRro.mkv"
] | 4
|
<answer>1</answer>
|
[
1
] | false
| 0
| 20,692.653134
| 281
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A man is on top of the roof talking about how to work up there using certain tools and putting them under the shingles. He was able to get one out from underneath and then begins to press down on some metal. He gets a hammer and begins to nail it in place and then he cuts off the excess. He lays down a new shingle to replace it over the metal piece he previously nailed in.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_mNM-JUC7ZEA.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_eBlYGGmeBY0.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_Oc8ACBiwIyE.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_LAicExwwM54.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_xf9iLflgRro.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A man in a kilt walks across a stage playing the bagpipe. The man on the stage stomps his feet. The camera zooms in on the man playing on the stage. The man finishes and raises his hand in the air and walks off the stage."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_bJ6SpcLM7GE.mp4",
"activitynet/videos/v_8v2ewQE-QK0.mp4",
"activitynet/videos/v_vynLNpomc30.mp4",
"activitynet/videos/v_s0N0PzdwVik.mp4",
"activitynet/videos/v_TVmuh_sR1KI.mp4"
] | 4
|
<answer>5</answer>
|
[
5
] | false
| 0
| 10,645.135235
| 283
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A man in a kilt walks across a stage playing the bagpipe. The man on the stage stomps his feet. The camera zooms in on the man playing on the stage. The man finishes and raises his hand in the air and walks off the stage.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_bJ6SpcLM7GE.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_8v2ewQE-QK0.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_vynLNpomc30.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_s0N0PzdwVik.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_TVmuh_sR1KI.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A foosball table is on a tiled floor. People are playing foosball indoors. The ball is knocked in a the goal on a foosball table."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_UuLBAMSmwgc.mp4",
"activitynet/videos/v_DP9hfhq8sro.mp4",
"activitynet/videos/v_m6C4SOxfNGQ.mkv",
"activitynet/videos/v_Zp9mSiw8Vkw.mp4",
"activitynet/videos/v_l4LFSd-7hxU.mp4"
] | 2
|
<answer>2</answer>
|
[
2
] | true
| 1
| 21,629.099589
| 282
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A foosball table is on a tiled floor. People are playing foosball indoors. The ball is knocked in a the goal on a foosball table.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_UuLBAMSmwgc.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_DP9hfhq8sro.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_m6C4SOxfNGQ.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_Zp9mSiw8Vkw.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_l4LFSd-7hxU.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A woman is standing outdoors with an accordion keyboard and pushes her sunglasses up. The woman then begins to play the accordion while swaying back and forth while an older man with a cigarette in his mouth is standing close to her watching and smiling. The woman stops playing, laughs and looks down."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_kBh_98QimD4.mp4",
"activitynet/videos/v_Rho3u46ZIEE.mp4",
"activitynet/videos/v_X6CpfuJLx0U.mp4",
"activitynet/videos/v_4R0fSNCWUo0.mp4",
"activitynet/videos/v_mYfo8LhPB5Y.mp4"
] | 2
|
<answer>2</answer>
|
[
2
] | true
| 1
| 21,284.042764
| 284
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A woman is standing outdoors with an accordion keyboard and pushes her sunglasses up. The woman then begins to play the accordion while swaying back and forth while an older man with a cigarette in his mouth is standing close to her watching and smiling. The woman stops playing, laughs and looks down.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_kBh_98QimD4.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_Rho3u46ZIEE.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_X6CpfuJLx0U.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_4R0fSNCWUo0.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_mYfo8LhPB5Y.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A man is seen standing in front of a group of people. The man then conducts the choir to sing together. The group continues to sing with one another while others watch on the side."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_4lxS8OJRsa8.mp4",
"activitynet/videos/v_AcLZk6JyXUM.mp4",
"activitynet/videos/v_yFJVEplkVHA.mp4",
"activitynet/videos/v_8FO4W-SBpxs.mp4",
"activitynet/videos/v_WOZbWqJMkRg.mp4"
] | 2
|
<answer>2</answer>
|
[
2
] | true
| 1
| 15,972.368799
| 285
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A man is seen standing in front of a group of people. The man then conducts the choir to sing together. The group continues to sing with one another while others watch on the side.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_4lxS8OJRsa8.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_AcLZk6JyXUM.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_yFJVEplkVHA.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_8FO4W-SBpxs.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_WOZbWqJMkRg.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A cheerleading team in teal, navy and white dance around a field in front of several girls and red and in pink tank tops. After milling around for a few minutes in a couple of clusters and talking with each other, they get into formation and do some of their routines. They seem to end with their impressive human pyramid, several girls on top of each other's shoulders. They hug and congratulate each other on a job well-done and acknowledge the girls seated waiting in their tank tops, and seem to get into another formation."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_M0sa3xWhFGo.mp4",
"activitynet/videos/v_7ih5UMIU7zE.mp4",
"activitynet/videos/v_EQPiYEvFmSo.mp4",
"activitynet/videos/v_m3yLm_dJU94.mp4",
"activitynet/videos/v_AISkvED80lU.mp4"
] | 1
|
<answer>1</answer>
|
[
1
] | true
| 1
| 20,492.135923
| 286
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A cheerleading team in teal, navy and white dance around a field in front of several girls and red and in pink tank tops. After milling around for a few minutes in a couple of clusters and talking with each other, they get into formation and do some of their routines. They seem to end with their impressive human pyramid, several girls on top of each other's shoulders. They hug and congratulate each other on a job well-done and acknowledge the girls seated waiting in their tank tops, and seem to get into another formation.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_M0sa3xWhFGo.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_7ih5UMIU7zE.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_EQPiYEvFmSo.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_m3yLm_dJU94.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_AISkvED80lU.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A young man sucks smoke through a hookah pipe. The boy blows smoke rings through the air. The man sucks back in the smoke ring floating in the air."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_u02UsNRxclU.mp4",
"activitynet/videos/v_zz3Mw8FMA70.mp4",
"activitynet/videos/v_2sbF8W0_bbg.mp4",
"activitynet/videos/v_CMTiL1ctmDs.mp4",
"activitynet/videos/v_i2X7z9ywHV8.mp4"
] | 4
|
<answer>5</answer>
|
[
5
] | false
| 0
| 14,168.723541
| 287
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A young man sucks smoke through a hookah pipe. The boy blows smoke rings through the air. The man sucks back in the smoke ring floating in the air.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_u02UsNRxclU.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_zz3Mw8FMA70.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_2sbF8W0_bbg.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_CMTiL1ctmDs.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_i2X7z9ywHV8.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "We see a rowing machine in a exercise room. A lady sits on the seat and pulls the cord of the rowing machine. We are shown the components of the machine. A lady grabs the handle bar of the machine. We see a lady push buttons on the display of the machine. A lady adjusts he paddle and straps her foot in. We see a lady put the machine away."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_sR3_5j8pUdI.mp4",
"activitynet/videos/v_9A9_sNvJ8zQ.mp4",
"activitynet/videos/v_qhnWJ4G5JMA.mp4",
"activitynet/videos/v_CHMk7efu1ro.mp4",
"activitynet/videos/v_e6mpdQ3BFhA.mp4"
] | 1
|
<answer>4</answer>
|
[
4
] | false
| 0
| 20,845.505456
| 288
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"We see a rowing machine in a exercise room. A lady sits on the seat and pulls the cord of the rowing machine. We are shown the components of the machine. A lady grabs the handle bar of the machine. We see a lady push buttons on the display of the machine. A lady adjusts he paddle and straps her foot in. We see a lady put the machine away.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_sR3_5j8pUdI.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_9A9_sNvJ8zQ.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_qhnWJ4G5JMA.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_CHMk7efu1ro.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_e6mpdQ3BFhA.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A woman with long curly hair speaks to the camera, she has a large red hairdryer with an attachment. She examines the parts of the dryer and shows off hair product. She uses it to dry her hair followed by a comb and shake. Credits and social media links end the video."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_UodvUEkuVig.mp4",
"activitynet/videos/v_hJf7uOUiEFo.mp4",
"activitynet/videos/v_wONwHYy59Tc.mp4",
"activitynet/videos/v_N2WxAkVh-C4.mp4",
"activitynet/videos/v_eTVzSwuCfd8.mp4"
] | 3
|
<answer>3</answer>
|
[
3
] | true
| 1
| 26,005.767357
| 289
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A woman with long curly hair speaks to the camera, she has a large red hairdryer with an attachment. She examines the parts of the dryer and shows off hair product. She uses it to dry her hair followed by a comb and shake. Credits and social media links end the video.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_UodvUEkuVig.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_hJf7uOUiEFo.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_wONwHYy59Tc.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_N2WxAkVh-C4.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_eTVzSwuCfd8.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "We see a colorful title shot. We see people in a gymnasium doing dance aerobics. The people kick their legs back and forth. The people perform the kicks again. We see the closing title on two screens."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_zmmiX3_TJ84.mp4",
"activitynet/videos/v_vL8Hy6lcnF8.mp4",
"activitynet/videos/v_gGai6uu5Yjs.mp4",
"activitynet/videos/v_n_9skH6xGeM.mp4",
"activitynet/videos/v_Ol7JKNItQC4.mp4"
] | 1
|
<answer>1</answer>
|
[
1
] | true
| 1
| 19,202.176658
| 291
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"We see a colorful title shot. We see people in a gymnasium doing dance aerobics. The people kick their legs back and forth. The people perform the kicks again. We see the closing title on two screens.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_zmmiX3_TJ84.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_vL8Hy6lcnF8.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_gGai6uu5Yjs.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_n_9skH6xGeM.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_Ol7JKNItQC4.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A group of balls are on a pool table being shot. Two men are shown playing a game of pool, then standing and talking with their sticks and drinking an energy drink. The man is shown interviewing numerous different people while talking about the drink. He wins a final game, his opponent in a neck brace."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_0LJ1mSpqGJg.mp4",
"activitynet/videos/v_5ptxyeHlcwM.mp4",
"activitynet/videos/v_ZZ71FIfxX-c.mp4",
"activitynet/videos/v_cdpPn-7R3GQ.mp4",
"activitynet/videos/v_le0tJsyuPks.mp4"
] | 4
|
<answer>4</answer>
|
[
4
] | true
| 1
| 28,540.114496
| 290
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A group of balls are on a pool table being shot. Two men are shown playing a game of pool, then standing and talking with their sticks and drinking an energy drink. The man is shown interviewing numerous different people while talking about the drink. He wins a final game, his opponent in a neck brace.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_0LJ1mSpqGJg.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_5ptxyeHlcwM.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_ZZ71FIfxX-c.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_cdpPn-7R3GQ.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_le0tJsyuPks.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A camera pans around a wooden floor and shows a person walking downstairs. The person runs their hands along a carpet and pushes it along the floor. The man nails down the carpet while still pushing it down and cutting the sides. The man shows off the finished carpet in the end."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_iUVz4A5oblQ.mp4",
"activitynet/videos/v__ye90Ou8SnE.mp4",
"activitynet/videos/v_KbEoaYhMZ6c.mp4",
"activitynet/videos/v_EInkc1uEX3c.mp4",
"activitynet/videos/v_CdjU2OZri4c.mp4"
] | 1
|
<answer>1</answer>
|
[
1
] | true
| 1
| 12,596.231767
| 292
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A camera pans around a wooden floor and shows a person walking downstairs. The person runs their hands along a carpet and pushes it along the floor. The man nails down the carpet while still pushing it down and cutting the sides. The man shows off the finished carpet in the end.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_iUVz4A5oblQ.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v__ye90Ou8SnE.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_KbEoaYhMZ6c.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_EInkc1uEX3c.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_CdjU2OZri4c.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A young Asian girl is kneeling down in the sand packing it with her hand into the container. She then flips her toy over and makes a sand castle."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_Cy2wqpjppy8.mp4",
"activitynet/videos/v_ZH8hnmjRDsI.mp4",
"activitynet/videos/v_skIP_U4EYDQ.mp4",
"activitynet/videos/v_aELu8QS8T54.mp4",
"activitynet/videos/v_tZzse87ICr8.mp4"
] | 3
|
<answer>3</answer>
|
[
3
] | true
| 1
| 10,818.684634
| 293
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A young Asian girl is kneeling down in the sand packing it with her hand into the container. She then flips her toy over and makes a sand castle.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_Cy2wqpjppy8.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_ZH8hnmjRDsI.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_skIP_U4EYDQ.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_aELu8QS8T54.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_tZzse87ICr8.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A woman is seen standing in front of a camera holding onto a rubix cube. She begins attempting to solve the puzzle by moving her hands around. She finishes the rubix cube and presents it to the camera."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_2Voht8wf3dQ.mp4",
"activitynet/videos/v_D0RDF1ez-8Y.mp4",
"activitynet/videos/v_MzNI-qdQfQc.mp4",
"activitynet/videos/v_HMIv7qpDmH0.mkv",
"activitynet/videos/v_12v5k4Z8lAE.mp4"
] | 4
|
<answer>3</answer>
|
[
3
] | false
| 0
| 13,727.514495
| 294
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A woman is seen standing in front of a camera holding onto a rubix cube. She begins attempting to solve the puzzle by moving her hands around. She finishes the rubix cube and presents it to the camera.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_2Voht8wf3dQ.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_D0RDF1ez-8Y.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_MzNI-qdQfQc.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_HMIv7qpDmH0.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_12v5k4Z8lAE.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A boy walks backwards on the lawn. The boy picks up the croquet stick and makes his shot. The boy smiles and heads towards the ball."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_F8K9WQfHth4.mp4",
"activitynet/videos/v_N9xp9VbpklQ.mp4",
"activitynet/videos/v_zwa44U585FE.mp4",
"activitynet/videos/v_uz91AvGxjbw.mp4",
"activitynet/videos/v_gA0m7YUH408.mp4"
] | 2
|
<answer>5</answer>
|
[
5
] | false
| 0
| 10,918.774391
| 295
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A boy walks backwards on the lawn. The boy picks up the croquet stick and makes his shot. The boy smiles and heads towards the ball.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_F8K9WQfHth4.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_N9xp9VbpklQ.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_zwa44U585FE.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_uz91AvGxjbw.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_gA0m7YUH408.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A woman is outside of what looks to be a restaurant cutting up something into small chunks on a white cutting board. Once finished,the bowl is placed over and it and the glad begins to cut onions and green peppers in the bowl. Someone has given her a phone and the lady is talking on the phone while placing the contents in the bowl. Finally,the woman puts all of the ingredients in a cup and serves it to the person standing at the counter."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_U40FhqwfBvs.mp4",
"activitynet/videos/v_ADUmfTuiDH8.mp4",
"activitynet/videos/v_kj8L5yu-fGs.mp4",
"activitynet/videos/v_5AG9Q5bF4pM.mp4",
"activitynet/videos/v_fJCkM6secVM.mp4"
] | 3
|
<answer>4</answer>
|
[
4
] | false
| 0
| 11,840.555592
| 298
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A woman is outside of what looks to be a restaurant cutting up something into small chunks on a white cutting board. Once finished,the bowl is placed over and it and the glad begins to cut onions and green peppers in the bowl. Someone has given her a phone and the lady is talking on the phone while placing the contents in the bowl. Finally,the woman puts all of the ingredients in a cup and serves it to the person standing at the counter.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_U40FhqwfBvs.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_ADUmfTuiDH8.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_kj8L5yu-fGs.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_5AG9Q5bF4pM.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_fJCkM6secVM.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A young man talks next a small tree in a squatting position. A man stands on back a toddler on the border of a swimming pool. Then, the man raise the toddler by the hands and throw her to the water. The toddler swims in the swimming poll, then she each a woman who hugs the toddler."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_M_WEOecjwLY.mkv",
"activitynet/videos/v_udpVICVTQrQ.mp4",
"activitynet/videos/v_yINX46xPRf0.mp4",
"activitynet/videos/v_DguywhRJ7ds.mp4",
"activitynet/videos/v_RjZ7jJBE1Qw.mp4"
] | 5
|
<answer>4</answer>
|
[
4
] | false
| 0
| 13,927.779635
| 297
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A young man talks next a small tree in a squatting position. A man stands on back a toddler on the border of a swimming pool. Then, the man raise the toddler by the hands and throw her to the water. The toddler swims in the swimming poll, then she each a woman who hugs the toddler.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_M_WEOecjwLY.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_udpVICVTQrQ.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_yINX46xPRf0.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_DguywhRJ7ds.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_RjZ7jJBE1Qw.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "The credits of the clip are shown. In a black and white, people are climbing the snow and pulling snow sleds. In a black and white, people walk down a tunnel. In a black and white, people ride an escalator with inflated tires. Individuals slide down the snow with inflated tires. The credits of the video are shown."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_5kmGgH4xFW0.mp4",
"activitynet/videos/v_vdq_xoRyxCU.mp4",
"activitynet/videos/v_eaI8My4pGq4.mp4",
"activitynet/videos/v_xnCw4tvy0uQ.mp4",
"activitynet/videos/v_fU-kGMQ68jg.mp4"
] | 2
|
<answer>2</answer>
|
[
2
] | true
| 1
| 21,108.008137
| 296
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"The credits of the clip are shown. In a black and white, people are climbing the snow and pulling snow sleds. In a black and white, people walk down a tunnel. In a black and white, people ride an escalator with inflated tires. Individuals slide down the snow with inflated tires. The credits of the video are shown.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_5kmGgH4xFW0.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_vdq_xoRyxCU.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_eaI8My4pGq4.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_xnCw4tvy0uQ.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_fU-kGMQ68jg.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.
System Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.
Decision rules (concise):
- Judge relevance ONLY from what is visible in the frames.
- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.
- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.
ABSOLUTE OUTPUT CONTRACT:
<answer> [a] > [b] > [c] > [d] > [e] </answer>
Constraints:
- Do NOT use per-candidate XML tags.
- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.
- Always output a total order in <answer>.
Query: "A woman jumps on the diving board and spins four times in the air, then dive in the water. Then, the woman runs and spins four times, and dive in the water. After, the woman stands backwards and stand on her hands and then spins four times."
Candidates:
[1] video:<video>
[2] video:<video>
[3] video:<video>
[4] video:<video>
[5] video:<video>
|
[
"activitynet/videos/v_EYIYohKR0Qo.mp4",
"activitynet/videos/v_AdP2aMo6OgY.mp4",
"activitynet/videos/v_bvnXdr-Hre4.mp4",
"activitynet/videos/v_D2JvqkKa-qM.mkv",
"activitynet/videos/v_y0Kio7VOk5o.mp4"
] | 2
|
<answer>1</answer>
|
[
1
] | false
| 0
| 12,131.727883
| 300
|
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "You will receive a Query and N candidates. Each candidate is a video. Your task is to identify the most relevant video to the Query. Output only the index of the most relevant video among the candidates. Your answer should be strictly inside <answer></answer> tags.\n\nSystem Prompt: You are RankLLM, a vision-only reranker. You will receive a Query and N candidates. Each candidate is a video described via frames.\n\nDecision rules (concise):\n- Judge relevance ONLY from what is visible in the frames.\n- Down-rank clear mismatches (wrong domain/scene/action). Consider temporal coherence.\n- Tie-breakers (in order): action visibility/close-up > temporal coherence > low occlusion/clutter > overall coverage.\n\nABSOLUTE OUTPUT CONTRACT:\n\n<answer> [a] > [b] > [c] > [d] > [e] </answer>\n\n\nConstraints:\n- Do NOT use per-candidate XML tags.\n- a,b,c,d,e are indices of the candidates. you should output the number in the place of a,b,c,d,e.\n- Always output a total order in <answer>.\n\nQuery: \"A woman jumps on the diving board and spins four times in the air, then dive in the water. Then, the woman runs and spins four times, and dive in the water. After, the woman stands backwards and stand on her hands and then spins four times.\"\n\nCandidates:\n[1] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_EYIYohKR0Qo.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[2] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_AdP2aMo6OgY.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[3] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_bvnXdr-Hre4.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[4] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_D2JvqkKa-qM.mp4",
"nframes": 48,
"total_pixels": 50176
}
},
{
"type": "text",
"text": "\n[5] video:",
"video_url": null
},
{
"type": "video_url",
"text": null,
"video_url": {
"url": "file:///hub_data3/seohyun/activitynet/videos/v_y0Kio7VOk5o.mp4",
"nframes": 48,
"total_pixels": 50176
}
}
]
}
] |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.