Download VideoX-Fun/output_debug/train_sample_full/sample-100-4.json from YFanwang/Backup: direct link, hf CLI and curl.
- Browser
- Download file 3.91 kB
-
https://huggingface.co/datasets/YFanwang/Backup/resolve/main/VideoX-Fun/output_debug/train_sample_full/sample-100-4.json
- Command line
-
hf download hf://datasets/YFanwang/Backup/VideoX-Fun/output_debug/train_sample_full/sample-100-4.json
-
curl -L -o sample-100-4.json https://huggingface.co/datasets/YFanwang/Backup/resolve/main/VideoX-Fun/output_debug/train_sample_full/sample-100-4.json
3.91 kB
| { | |
| "prompt": [ | |
| "The video is a news segment from NBC News, featuring a woman with blonde hair and a black dress. She is speaking about a 4-day work week experiment that shows productivity is equal to a 5-day work week. The woman is standing in front of a brick wall with a wooden panel, and there are other people in the background. The style of the video is informative and professional, with a focus on delivering news content. The woman's expression is serious, and she is looking directly at the camera. The background is simple and uncluttered, allowing the viewer to focus on the woman and her message. The video is likely intended for a general audience, and the content is likely to be of interest to those who are interested in work-life balance and productivity." | |
| ], | |
| "video_ids": [ | |
| "Ig7ast24Em8_30_144to280" | |
| ], | |
| "question": [ | |
| "You are a careful video forensics assistant. You are given two videos: a test video (the first 2 frames) which is ai-generated and a real video (the last 10 frames). The test video is generated by the caption of the real video. The caption is \"The video is a news segment from NBC News, featuring a woman with blonde hair and a black dress. She is speaking about a 4-day work week experiment that shows productivity is equal to a 5-day work week. The woman is standing in front of a brick wall with a wooden panel, and there are other people in the background. The style of the video is informative and professional, with a focus on delivering news content. The woman's expression is serious, and she is looking directly at the camera. The background is simple and uncluttered, allowing the viewer to focus on the woman and her message. The video is likely intended for a general audience, and the content is likely to be of interest to those who are interested in work-life balance and productivity.\"\nYour task is evaluating whether the test video shows physics-related defects. You should compared both videos and use the provided caption as high-level intent. Focus on physical plausibility, not style or aesthetics.\nYou need to analyze it in these aspects:\n- Liquid flow irregularity \u2013 e.g., non-inertial or discontinuous flow, volume popping, gravity-inconsistent motion, impossible splashes.\n- Abnormal object deformation \u2013 e.g., rigid objects bending/stretching without cause, topology changes (parts merging/splitting).\n- Abnormal texture/material change \u2013 e.g., surface turns matte\u2192glossy without cause, texture swimming/flicker detached from geometry.\n- Abnormal motion \u2013 e.g., inertia/acceleration violations, teleporting, time reversals, jitter not explained by camera motion.\n- Unnatural interpenetration \u2013 e.g., objects passing through each other or ground, missing collisions/contacts.\nIf the prompt doesn't have related physical aspects, you should answer 'No'.\nOutput strictly as JSON with this schema (no extra text):\n{\n \"liquid_flow_irregularity\": \"Yes/No\",\n \"abnormal_deformation\": \"Yes/No\",\n \"abnormal_texture_change\": \"Yes/No\", \n \"abnormal_motion\": \"Yes/No\", \n \"unnatural_interpenetration\": \"Yes/No\",\n}\n" | |
| ], | |
| "gt_answer": [ | |
| "```json\n{\n \"liquid_flow_irregularity\": \"No\",\n \"abnormal_deformation\": \"No\",\n \"abnormal_texture_change\": \"No\",\n \"abnormal_motion\": \"No\",\n \"unnatural_interpenetration\": \"No\"\n}\n```" | |
| ], | |
| "pred_answer": [ | |
| "json\n{\n \"liquid_flow_irregularity\": \"No\",\n \"abnormal_deformation\": \"No\",\n \"abnormal_texture_change\": \"No\",\n \"abnormal_motion\": \"No\",\n \"unnatural_interpenetration\": \"No\"\n}\n```<|im_end|>\n<|im_start|>", | |
| " \"has_obvious_defect\"\": No,\n \"\"dominant_issue\"\": \"\"none\"\",\n detail_loss_vs_reference\"\": NoNo\"\",\n evidence\"\": [}<|im_end|>\n<|im_start|>" | |
| ] | |
| } |