Model accuracy in validate function and test.py varies for the same model and same set of images.

[Aratrik]

No description provided.

[Aratrik]

Hi Thanks for providing the code. I was testing the code for a pspNet-101 model I trained from scratch. When I checked the mIoU score of the model on validation images, the validate function score and test.py function score did not match for the same model. Is there any insight why this might happen. In validate I am using a batch size of 2 with no distributed inference.