PKU-YuanGroup / Video-LLaVA

【EMNLP 2024🔥】Video-LLaVA: Learning United Visual Representation by Alignment Before Projection
https://arxiv.org/pdf/2311.10122.pdf
Apache License 2.0
3.02k stars 220 forks source link

ERROR opening+moov atom not found+mmco: unref short failure #163

Open Frank-Dg opened 6 months ago

Frank-Dg commented 6 months ago

hello,thanks for your codes!I train the model during finetune,but i found those errors: 1、 89%|████████▉ | 5307/5979 [15:40:54<2:01:44, 10.87s/it][mov,mp4,m4a,3gp,3g2,mj2 @ 0xf78b0000] moov atom not found [02:44:49] /github/workspace/src/video/video_reader.cc:83: ERROR opening: /data/video_llava_data/train_data/videochatgpt_tune/v_MvZFYjs80Y4.mp4, Invalid data found when processing input Error with Error reading /data/video_llava_data/train_data/videochatgpt_tune/v_MvZFYjs80Y4.mp4... 2、89%|████████▉ | 5308/5979 [15:41:04<1:59:05, 10.65s/it][h264 @ 0xa5268c0] mmco: unref short failure [h264 @ 0xd682040] mmco: unref short failure

q1:Upon checking the video, I found that v_MvZFYjs80Y4.mp4 is 0 byte. I downloaded the dataset using the link you provided. Have you encountered this problem before? q2:While the model could technically continue training, is it appropriate to simply discard these unopenable video files?

haopenghui commented 5 months ago

我也遇到了这个问题,请问是怎么解决的?

btekin commented 5 months ago

same issue, can we simply ignore these files with such errors to continue training and get successful convergence?