Eval mean loss increasing in sem_seg

[uzair-mehmood-storex]

Hello, I really appreciate your work, it does help me a lot in my research. Well, I've prepared my own dataset of semantic segmentation with different number of classes. I've changed, configuration of network for these number of classes and trained vanilla point net. I've run several attempts but, no matter, how I distribute my dataset in train and eval, train mean loss always decreases but, eval mean loss fluctuates, but not decrease. I even visualize the output of trained model on eval and test dataset, but model perform really bad on them, on other hand it perform really well on training dataset. Could you guide me further on this issue.

[Swaraj-72]

Hello @uzair-mehmood-storex , have you found any solution for this issue? I am having similar troubles while dealing with a custom dataset.