There is a memory leak in the training code of AdaPoinTr_PCN model when I train my own model, and my training memory of more than 2200epoch can occupy about 40G+.By the memory, I don't mean the graphics memory of the graphics card, but the memory of the computer.
Hey @cckamiya, it appears that we're saving the picture during each validation here. If you believe the visualization in TFBorad isn't needed, go ahead and comment out this section of the code to see if it resolves the issue.
There is a memory leak in the training code of AdaPoinTr_PCN model when I train my own model, and my training memory of more than 2200epoch can occupy about 40G+.By the memory, I don't mean the graphics memory of the graphics card, but the memory of the computer.