Datasets:
happy8825
/
qweninstruct_tuned

Dataset card Data Studio Files
xet
Community
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: "People are playing a game of pool in a room. People are standing by a wall watching them. People are sitting at a bar behind them watching them." Candidates: [1] video:<video> [2] video:<video> [3] video:<video> [4] video:<video> [5] video:<video>
[ "activitynet/videos/v_P5Y-b-lcBs0.mp4", "activitynet/videos/v_8yeUJm0Pl24.mp4", "activitynet/videos/v_0bzSBV3jHIY.mp4", "activitynet/videos/v_R_TRpIHkgMs.mp4", "activitynet/videos/v_fjN9Qe237bw.mp4" ]
1
<answer>3</answer>
[ 3 ]
false
0
25,370.303678
4,899
[ { "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 a game of pool in a room. People are standing by a wall watching them. People are sitting at a bar behind them watching them.\"\n\nCandidates:\n[1] video:", "video_url": null }, { "type": "video_url", "text": null, "video_url": { "url": "file:///hub_data3/seohyun/activitynet/videos/v_P5Y-b-lcBs0.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_8yeUJm0Pl24.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_0bzSBV3jHIY.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_R_TRpIHkgMs.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_fjN9Qe237bw.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 snowboard on a steep mountain going down without falling. Other men snowboard down the mountain falling from very high and land stand on the mountain. A man flips in the air, then lands on the snow to continue skiing down the mountain with deep snow. Other people snowboard on the mountain going down the slopes. A person drives a snow motorcycle on the mountains. After, men continues snowboard on the high mountain with deep snow." Candidates: [1] video:<video> [2] video:<video> [3] video:<video> [4] video:<video> [5] video:<video>
[ "activitynet/videos/v_PAiJNr97C6g.mp4", "activitynet/videos/v_zOGg5-Mll4o.mp4", "activitynet/videos/v_jF33TElZc_Y.mp4", "activitynet/videos/v_KoHzXi7Usl8.mp4", "activitynet/videos/v_dUzqM38vwPI.mp4" ]
3
<answer>4</answer>
[ 4 ]
false
0
19,265.744597
4,901
[ { "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 snowboard on a steep mountain going down without falling. Other men snowboard down the mountain falling from very high and land stand on the mountain. A man flips in the air, then lands on the snow to continue skiing down the mountain with deep snow. Other people snowboard on the mountain going down the slopes. A person drives a snow motorcycle on the mountains. After, men continues snowboard on the high mountain with deep snow.\"\n\nCandidates:\n[1] video:", "video_url": null }, { "type": "video_url", "text": null, "video_url": { "url": "file:///hub_data3/seohyun/activitynet/videos/v_PAiJNr97C6g.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_zOGg5-Mll4o.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_jF33TElZc_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_KoHzXi7Usl8.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_dUzqM38vwPI.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 stands against a wall playing guitar and singing. Then, the man stops playing and then talks." Candidates: [1] video:<video> [2] video:<video> [3] video:<video> [4] video:<video> [5] video:<video>
[ "activitynet/videos/v_a8dUtKcAunw.mp4", "activitynet/videos/v_W1krUTxgsMc.mp4", "activitynet/videos/v_jUmfhYsA5r0.mp4", "activitynet/videos/v_sARnRvNdl-Y.mp4", "activitynet/videos/v_akwJwcvfjLA.mp4" ]
3
<answer>2</answer>
[ 2 ]
false
0
19,196.450438
4,902
[ { "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 stands against a wall playing guitar and singing. Then, the man stops playing and then talks.\"\n\nCandidates:\n[1] video:", "video_url": null }, { "type": "video_url", "text": null, "video_url": { "url": "file:///hub_data3/seohyun/activitynet/videos/v_a8dUtKcAunw.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_W1krUTxgsMc.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_jUmfhYsA5r0.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_sARnRvNdl-Y.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_akwJwcvfjLA.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 Justin Bieber is playing drums in a studio. Two of his friends are watching while recording on their phone. Justin ends the drum solo by throwing his drum sticks to the side. His friends applaud his playing." Candidates: [1] video:<video> [2] video:<video> [3] video:<video> [4] video:<video> [5] video:<video>
[ "activitynet/videos/v_Sd08rsPTroE.mp4", "activitynet/videos/v_4ACqWG_p1bI.mp4", "activitynet/videos/v_RnwidjJiDEE.mp4", "activitynet/videos/v_xQH8YS2_NxY.mkv", "activitynet/videos/v_NNqghz7Fd0M.mp4" ]
4
<answer>4</answer>
[ 4 ]
true
1
27,430.838737
4,903
[ { "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 Justin Bieber is playing drums in a studio. Two of his friends are watching while recording on their phone. Justin ends the drum solo by throwing his drum sticks to the side. His friends applaud his playing.\"\n\nCandidates:\n[1] video:", "video_url": null }, { "type": "video_url", "text": null, "video_url": { "url": "file:///hub_data3/seohyun/activitynet/videos/v_Sd08rsPTroE.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_4ACqWG_p1bI.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_RnwidjJiDEE.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_xQH8YS2_NxY.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_NNqghz7Fd0M.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 kid is in the yard putting the ball on the side of the fence. An adult came and start to cut grass with the kid. The man moved the goal net on the side and continue to cut the grass using the grass cutter, then he move the trampoline and move cut the grass as the boy is playing on the side." Candidates: [1] video:<video> [2] video:<video> [3] video:<video> [4] video:<video> [5] video:<video>
[ "activitynet/videos/v_4mBVik8dq_w.mp4", "activitynet/videos/v_8RMrbKCQheM.mp4", "activitynet/videos/v_Sjx7K9Ybx9Q.mp4", "activitynet/videos/v_JSqJmZPqDy8.mp4", "activitynet/videos/v_oy1XjDer7o4.mp4" ]
5
<answer>5</answer>
[ 5 ]
true
1
15,887.665132
4,904
[ { "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 kid is in the yard putting the ball on the side of the fence. An adult came and start to cut grass with the kid. The man moved the goal net on the side and continue to cut the grass using the grass cutter, then he move the trampoline and move cut the grass as the boy is playing 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_4mBVik8dq_w.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_8RMrbKCQheM.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_Sjx7K9Ybx9Q.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_JSqJmZPqDy8.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_oy1XjDer7o4.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 yard and leads into a man raking up leaves and putting them into bags. The man continues stuffing leads into a bag as well as raking leaves all around the yard." Candidates: [1] video:<video> [2] video:<video> [3] video:<video> [4] video:<video> [5] video:<video>
[ "activitynet/videos/v_RblRzlmSFak.mp4", "activitynet/videos/v_HpjomKhpIdk.mp4", "activitynet/videos/v_2Sev8z4P7pE.mp4", "activitynet/videos/v_5WqnKjOz1z4.mkv", "activitynet/videos/v_gY0qgtM0Gt8.mp4" ]
4
<answer>3</answer>
[ 3 ]
false
0
9,802.092721
4,905
[ { "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 yard and leads into a man raking up leaves and putting them into bags. The man continues stuffing leads into a bag as well as raking leaves all around the yard.\"\n\nCandidates:\n[1] video:", "video_url": null }, { "type": "video_url", "text": null, "video_url": { "url": "file:///hub_data3/seohyun/activitynet/videos/v_RblRzlmSFak.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_HpjomKhpIdk.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_2Sev8z4P7pE.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_5WqnKjOz1z4.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_gY0qgtM0Gt8.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 sitting before a large mat with another person laying in the background. The man then runs down the mats on stilts and jumps over a large mat." Candidates: [1] video:<video> [2] video:<video> [3] video:<video> [4] video:<video> [5] video:<video>
[ "activitynet/videos/v_m--b-Ltjm_Y.mp4", "activitynet/videos/v_DV1ITGBfo5w.mp4", "activitynet/videos/v_cgPt46YiXNo.mp4", "activitynet/videos/v_2AQg1DDVYHI.mp4", "activitynet/videos/v_E2Vd-sOC_ik.mp4" ]
3
<answer>4</answer>
[ 4 ]
false
0
11,362.100936
4,907
[ { "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 sitting before a large mat with another person laying in the background. The man then runs down the mats on stilts and jumps over a large mat.\"\n\nCandidates:\n[1] video:", "video_url": null }, { "type": "video_url", "text": null, "video_url": { "url": "file:///hub_data3/seohyun/activitynet/videos/v_m--b-Ltjm_Y.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_DV1ITGBfo5w.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_cgPt46YiXNo.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_2AQg1DDVYHI.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_E2Vd-sOC_ik.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 red title screen is shown where a man is tossing food in a frying pan. The a male and female are shown on a television production stage behind an island with cooking ingredients. The chef then begins to describe the food in front of him and cuts the food to go into the pan. Once completed, he cuts eggs in half and cuts up small green things as the woman assists him by turning the food in the pan." Candidates: [1] video:<video> [2] video:<video> [3] video:<video> [4] video:<video> [5] video:<video>
[ "activitynet/videos/v_O1z0Q-3OUg0.mp4", "activitynet/videos/v_9PP5_HGpu4c.mp4", "activitynet/videos/v_h3thb-S-3L8.mp4", "activitynet/videos/v_TscC5kgurqY.mp4", "activitynet/videos/v_tTkavaWq0QM.mp4" ]
2
<answer>2</answer>
[ 2 ]
true
1
14,932.562765
4,908
[ { "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 red title screen is shown where a man is tossing food in a frying pan. The a male and female are shown on a television production stage behind an island with cooking ingredients. The chef then begins to describe the food in front of him and cuts the food to go into the pan. Once completed, he cuts eggs in half and cuts up small green things as the woman assists him by turning the food in the pan.\"\n\nCandidates:\n[1] video:", "video_url": null }, { "type": "video_url", "text": null, "video_url": { "url": "file:///hub_data3/seohyun/activitynet/videos/v_O1z0Q-3OUg0.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_9PP5_HGpu4c.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_h3thb-S-3L8.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_TscC5kgurqY.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_tTkavaWq0QM.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 at a table. She is knitting with pink yarn and needles. She talks as she continues to knit." Candidates: [1] video:<video> [2] video:<video> [3] video:<video> [4] video:<video> [5] video:<video>
[ "activitynet/videos/v_iBefG1qFbsE.mp4", "activitynet/videos/v_mglEC2-MH14.mp4", "activitynet/videos/v_tj0sI8M3tro.mkv", "activitynet/videos/v_6VUsbs84lCc.mp4", "activitynet/videos/v_ItukN-TWrJM.mkv" ]
4
<answer>4</answer>
[ 4 ]
true
1
20,495.303624
4,906
[ { "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 at a table. She is knitting with pink yarn and needles. She talks as she continues to knit.\"\n\nCandidates:\n[1] video:", "video_url": null }, { "type": "video_url", "text": null, "video_url": { "url": "file:///hub_data3/seohyun/activitynet/videos/v_iBefG1qFbsE.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_mglEC2-MH14.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_tj0sI8M3tro.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_6VUsbs84lCc.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_ItukN-TWrJM.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 picture of a little girl in a ballerina skirt is shown. Several little girls in ballerina skirts are dancing on a stage. They put their hands above their heads and run off stage." Candidates: [1] video:<video> [2] video:<video> [3] video:<video> [4] video:<video> [5] video:<video>
[ "activitynet/videos/v_bmIWsU8sNlw.mkv", "activitynet/videos/v_VDYSVR0HbpM.mp4", "activitynet/videos/v_UYe6JGaUZzg.mp4", "activitynet/videos/v_QzbZxKJ-YBY.mp4", "activitynet/videos/v_T8ae3_Pm5eE.mkv" ]
1
<answer>3</answer>
[ 3 ]
false
0
16,084.361437
4,909
[ { "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 picture of a little girl in a ballerina skirt is shown. Several little girls in ballerina skirts are dancing on a stage. They put their hands above their heads and run off stage.\"\n\nCandidates:\n[1] 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 } }, { "type": "text", "text": "\n[2] 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[3] video:", "video_url": null }, { "type": "video_url", "text": null, "video_url": { "url": "file:///hub_data3/seohyun/activitynet/videos/v_UYe6JGaUZzg.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_QzbZxKJ-YBY.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_T8ae3_Pm5eE.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 racatac is shown in a white room. a racatac is shown in a commecial showing the prices. different racatacs in sizes and models are shown in a room." Candidates: [1] video:<video> [2] video:<video> [3] video:<video> [4] video:<video> [5] video:<video>
[ "activitynet/videos/v_D7WhCBcddSA.mp4", "activitynet/videos/v_P4PQ5tC3gX8.mp4", "activitynet/videos/v_fxgbk_Kk4Rw.mp4", "activitynet/videos/v_cGoj8xGxrG0.mp4", "activitynet/videos/v_96c3BWVDoq4.mp4" ]
4
<answer>4</answer>
[ 4 ]
true
1
17,187.080819
4,910
[ { "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 racatac is shown in a white room. a racatac is shown in a commecial showing the prices. different racatacs in sizes and models are shown in a room.\"\n\nCandidates:\n[1] video:", "video_url": null }, { "type": "video_url", "text": null, "video_url": { "url": "file:///hub_data3/seohyun/activitynet/videos/v_D7WhCBcddSA.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_P4PQ5tC3gX8.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_fxgbk_Kk4Rw.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_cGoj8xGxrG0.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_96c3BWVDoq4.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 inside a building and outside. We see several scenes of a woman pulling a bow back from an archery set. She shoots the arrow toward a target." Candidates: [1] video:<video> [2] video:<video> [3] video:<video> [4] video:<video> [5] video:<video>
[ "activitynet/videos/v_Ht9WSqhFD34.mp4", "activitynet/videos/v_NsqW8ZwYDEk.mkv", "activitynet/videos/v_ur873jaQO3A.mkv", "activitynet/videos/v_x0PE_98UO3s.mp4", "activitynet/videos/v_WRc1Jv1j3nk.mp4" ]
5
<answer>4</answer>
[ 4 ]
false
0
19,076.333187
4,911
[ { "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 inside a building and outside. We see several scenes of a woman pulling a bow back from an archery set. She shoots the arrow toward a target.\"\n\nCandidates:\n[1] video:", "video_url": null }, { "type": "video_url", "text": null, "video_url": { "url": "file:///hub_data3/seohyun/activitynet/videos/v_Ht9WSqhFD34.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_NsqW8ZwYDEk.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_ur873jaQO3A.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_x0PE_98UO3s.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_WRc1Jv1j3nk.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 washing her hands. She lights a candle in a room. She is sitting down knitting. She is walking into a building. She sits down and talks with another lady. They all begin knitting sitting next to each other." Candidates: [1] video:<video> [2] video:<video> [3] video:<video> [4] video:<video> [5] video:<video>
[ "activitynet/videos/v_Mx6Gt14tnmY.mp4", "activitynet/videos/v_gEYutYXODs4.mp4", "activitynet/videos/v_TRXLUcm2CuQ.mp4", "activitynet/videos/v_w-KZEq6JhnQ.mp4", "activitynet/videos/v_sS-KyhAzeUY.mkv" ]
2
<answer>2</answer>
[ 2 ]
true
1
18,586.339273
4,912
[ { "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 washing her hands. She lights a candle in a room. She is sitting down knitting. She is walking into a building. She sits down and talks with another lady. They all begin knitting sitting next to each other.\"\n\nCandidates:\n[1] video:", "video_url": null }, { "type": "video_url", "text": null, "video_url": { "url": "file:///hub_data3/seohyun/activitynet/videos/v_Mx6Gt14tnmY.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_gEYutYXODs4.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_TRXLUcm2CuQ.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_w-KZEq6JhnQ.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_sS-KyhAzeUY.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 puppy is rubbing his face in someone's lap. He chews and licks his paw as he cuddles. He looks up at the camera slightly." Candidates: [1] video:<video> [2] video:<video> [3] video:<video> [4] video:<video> [5] video:<video>
[ "activitynet/videos/v_-MFzpFMdWZs.mp4", "activitynet/videos/v_K5wPwCFVkhU.mp4", "activitynet/videos/v_CuZpm0Il6YM.mp4", "activitynet/videos/v_RN2QwhcAsUY.mp4", "activitynet/videos/v_UPwDuuYlLfQ.mp4" ]
2
<answer>2</answer>
[ 2 ]
true
1
15,811.700049
4,913
[ { "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 puppy is rubbing his face in someone's lap. He chews and licks his paw as he cuddles. He looks up at the camera slightly.\"\n\nCandidates:\n[1] video:", "video_url": null }, { "type": "video_url", "text": null, "video_url": { "url": "file:///hub_data3/seohyun/activitynet/videos/v_-MFzpFMdWZs.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_K5wPwCFVkhU.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_CuZpm0Il6YM.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_RN2QwhcAsUY.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_UPwDuuYlLfQ.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 child is seen jumping on a set of monkey bars and moving along the bars. He climbs back towards the beginning and meets another child waiting." Candidates: [1] video:<video> [2] video:<video> [3] video:<video> [4] video:<video> [5] video:<video>
[ "activitynet/videos/v_0vQs3ztG7vg.mp4", "activitynet/videos/v_iEZgExTrv70.mp4", "activitynet/videos/v_bHAzuAnnvcU.mp4", "activitynet/videos/v_m16Cn9VA3Lk.mp4", "activitynet/videos/v_FIw076A69Oc.mp4" ]
1
<answer>1</answer>
[ 1 ]
true
1
13,504.925258
4,914
[ { "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 child is seen jumping on a set of monkey bars and moving along the bars. He climbs back towards the beginning and meets another child waiting.\"\n\nCandidates:\n[1] video:", "video_url": null }, { "type": "video_url", "text": null, "video_url": { "url": "file:///hub_data3/seohyun/activitynet/videos/v_0vQs3ztG7vg.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_iEZgExTrv70.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_bHAzuAnnvcU.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_m16Cn9VA3Lk.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_FIw076A69Oc.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 young men holds axes in the woods. Then, the young men splits logs in the woods." Candidates: [1] video:<video> [2] video:<video> [3] video:<video> [4] video:<video> [5] video:<video>
[ "activitynet/videos/v_dcEdjqyHj8M.mp4", "activitynet/videos/v_QMCHIR3nDLs.mkv", "activitynet/videos/v_iosb2TdQ7yY.mp4", "activitynet/videos/v_siGEHA6fs80.mp4", "activitynet/videos/v_PKYg6_rs3LQ.mp4" ]
5
<answer>5</answer>
[ 5 ]
true
1
14,749.123084
4,915
[ { "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 young men holds axes in the woods. Then, the young men splits logs in the woods.\"\n\nCandidates:\n[1] video:", "video_url": null }, { "type": "video_url", "text": null, "video_url": { "url": "file:///hub_data3/seohyun/activitynet/videos/v_dcEdjqyHj8M.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_QMCHIR3nDLs.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_iosb2TdQ7yY.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_siGEHA6fs80.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_PKYg6_rs3LQ.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 several shots of a mountain and shows people strapping rope onto themselves. A couple people are then seen climbing up large rocks while another waits at the bottom. They show more areas of the nearby location as well as more people climbing up the rock and riding around the area." Candidates: [1] video:<video> [2] video:<video> [3] video:<video> [4] video:<video> [5] video:<video>
[ "activitynet/videos/v_0HhNhRExwSQ.mp4", "activitynet/videos/v_5nOc03oiFvk.mp4", "activitynet/videos/v_J76bFZWXHFY.mp4", "activitynet/videos/v_VXCV3KUtCdk.mp4", "activitynet/videos/v_ICBrXUuwvgg.mp4" ]
2
<answer>4</answer>
[ 4 ]
false
0
20,894.835959
4,916
[ { "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 several shots of a mountain and shows people strapping rope onto themselves. A couple people are then seen climbing up large rocks while another waits at the bottom. They show more areas of the nearby location as well as more people climbing up the rock and riding around the area.\"\n\nCandidates:\n[1] video:", "video_url": null }, { "type": "video_url", "text": null, "video_url": { "url": "file:///hub_data3/seohyun/activitynet/videos/v_0HhNhRExwSQ.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_5nOc03oiFvk.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_J76bFZWXHFY.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_VXCV3KUtCdk.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_ICBrXUuwvgg.mp4", "nframes": 48, "total_pixels": 50176 } } ] } ]