Closed Zhenghongwei0929 closed 8 months ago
Thanks for your comments! We did not encounter this issue, and what do you mean by "stuck"? Did training completely stop or something?
Thanks for your replying! The training can still go on but it costs too much time for loading data every k iterations. I have checked that it is caused by the dataloader. But my server has abundant CPU and memory resources.
This is how long it takes for each iteration to load data.
Could you please provide a code snippet that produces these statistics (as well as the setting for training if you made any modifications)? I may try to look into this issue and check if this is the case for me.
I didn't change any code; I just counted the time taken for each iteration of next(batch). I noticed that I just now according to the guidance dealing with data in the H36M-Toolbox, not according to the ContextPose/MVN/datasets/human36m_preprocessing the steps in processing data, whether this will have effect? Thanks again for your reply!
We used the images produced by H36M-Toolbox but made some modifications to the original script to generate labels. Please follow our code to generate the labels! Actually, we also provided our pre-processed labels here, you can try to use them.
THX! I will try to process the data using the script in human36m_preprocessing.
Have you already solved this problem? I also encountered this issue, and I generated the labels using the author's script.
No, I still haven't solved this problem.
Thanks for your excellent work. Currently, data loads are stuck after every k iteration, where k is related to the number of num_workers I set for the data loader. I have tried this problem on different servers. May I ask what causes it?