search

fesvhtr / CUVA

[CVPR 2024] Official repository of the paper "Uncovering What, Why and How: A Comprehensive Benchmark for Causation Understanding of Video Anomaly"
29 stars 0 forks source link

IndexError: index 976 is out of bounds for dimension 0 with size 662 #3

Open happyrenxiaozhao opened 3 weeks ago

happyrenxiaozhao commented 3 weeks ago

Hello, thank you for your work! I recently tried to reproduce your code, but the following error occurred: IndexError: index 976 is out of bounds for dimension 0 with size 662

Here are some details I have, I hope you can help me out! 截图 2024-08-19 20-52-39 截图 2024-08-19 20-53-28

happyrenxiaozhao commented 3 weeks ago

@fesvhtr I am looking forward to your reply

happyrenxiaozhao commented 3 weeks ago

@fesvhtr I debugged the code and found the place where the error was reported. In line 106 of the video_chat.py file, the value video_start_token_pos + num_patches + 1 exceeds the index range. How should I modify it? 我调试了代码，并且找到了报错的地方，在video_chat.py这个文件的106行，video_start_token_pos + num_patches + 1这个值超出了索引范围，我应该怎么进行修改呢？ 截图 2024-08-20 11-20-42

Dulpy commented 3 weeks ago

It seems that the errors happened during the concatenation of the video tokens and text tokens, Here are some suggestions for you to check:

  1. Check the number of frames you extracted from the video
  2. Check the ModelArguments: mm_use_vid_start_end
  3. Check the video input, as we found that nearly 10 videos tested were not available, and we'll fix it in the next version And if it doesn't work, we will be releasing a new version of the model in the next few days, we'll be glad to invite you to try the new one.
happyrenxiaozhao commented 3 weeks ago

It seems that the errors happened during the concatenation of the video tokens and text tokens, Here are some suggestions for you to check:

  1. Check the number of frames you extracted from the video
  2. Check the ModelArguments: mm_use_vid_start_end
  3. Check the video input, as we found that nearly 10 videos tested were not available, and we'll fix it in the next version And if it doesn't work, we will be releasing a new version of the model in the next few days, we'll be glad to invite you to try the new one.

Thank you for your answer, I am very happy to look forward to your new version, and hope to receive your notice in time!