How can you get the meta files: .tex > Thanks for your feedback. You can set trainer.no_partial_bn = True if batch size >= 6 in each gpu and retry it, this will not affect the accuracy. That module exists some bug with distributed training, we will fix it quickly. about x-temporal

[deepcs233]

Maybe need to check the img_prefix in your config yaml, and make sure img_prefix % num_id can get the correct file name. The default setting is image_{:05d}.jpg .
The next thing is the video_source in the yaml: if your training data's type is video, you may need to set it as True.