Overfitting in 2D training.

[chaolongy]

Dear Mr. Jaritz, When I was reproducing the paper, there was an overfitting phenomenon when training a 2D network. The mIOU of 2D image semantic segmentation was only about 50%, and the visual results of inference were poor. Have you encountered this situation? Is there a better solution? I look forward to your reply.

[maxjaritz]

Hi, It is possible that overfitting occurs. You can try other data augmentation techniques if you want to improve.

Best, Max