HaoyuWu556

> 建议代码里改成按只读取视频中被sample的几帧，而不是全部加载。 谢谢 请问有文档写过数据集的采样逻辑吗 感觉这部分代码不是很好懂 也不太好找到改的地方

I think I have solved this issue by rewrite `read_video_cv2`

Moreover, I found there are videos more than 5k frames which is too much for memory. Maybe some pre-cutting would also work for this issue.

> > read_video_cv2 > > could you please share the codes of read_video_cv2? I meet a similiar problem Sorry, I still got deadlock or dataloader issue by my function. After...

> > > > read_video_cv2 > > > > > > > > > could you please share the codes of read_video_cv2? I meet a similiar problem > > >...

Thanks for your insight but I was wondering these are all work around. Finally I will need trianing on long videos.

torchrun --standalone --nproc_per_node 8 scripts/train.py \ configs/opensora-v1-2/train/stage1.py --data-path YOUR_CSV_PATH --ckpt-path YOUR_PRETRAINED_CKPT 训练环境是luchen的官方环境，训练代码用的是这个

This issue has been solved by loading from huggingface checkpoint. However loading from Opensora mirror from luchen cloud is still not working.

> > This issue has been solved by loading from huggingface checkpoint. However loading from Opensora mirror from luchen cloud is still not working. > > Can you explain in...

你的机器和环境是什么样的呀