search

OpenGVLab / VideoMamba

VideoMamba: State Space Model for Efficient Video Understanding
https://arxiv.org/abs/2403.06977
Apache License 2.0
660 stars 47 forks source link

float division by zero #36

Closed Sine7812 closed 2 months ago

Sine7812 commented 2 months ago

Traceback (most recent call last): File "/mnt/e/code/VideoMamba-main/videomamba/video_sm/run_class_finetuning.py", line 715, in main(opts, ds_init) File "/mnt/e/code/VideoMamba-main/videomamba/video_sm/run_class_finetuning.py", line 632, in main train_stats = train_one_epoch( File "/mnt/e/code/VideoMamba-main/videomamba/video_sm/engines/engine_for_finetuning.py", line 48, in train_one_epoch for data_iterstep, (samples, targets, , _) in enumerate(metric_logger.log_every(data_loader, print_freq, header)): File "/mnt/e/code/VideoMamba-main/videomamba/video_sm/utils.py", line 164, in log_every header, total_time_str, total_time / len(iterable))) ZeroDivisionError: float division by zero Has anyone encountered this problem before? How can I solve it? Thank you very much.

Andy1621 commented 2 months ago

Hi! Please check whether the dataloader is well prepared. The iterable may be null.

Sine7812 commented 2 months ago

Thank you for your reply. I have solved this problem. Since I just wanted to test whether the code could run, but the K400 dataset is too large, I only prepared a few video files and created a small dataset according to the author's format. However, when I tried distributed training, due to the relatively small size of the dataset and the possible inappropriate configuration of the batch size, the DataLoader may not have generated any batches during actual operation. After adjusting these settings, I successfully solved the problem.