Jeya Maria Jose

It can be changed in the code.

Yes, with the current code setup, this network needs fixed image size for all images in dataset as input.

You should remove those lines with mask if you are converting it to multi-class problem. Your ground truth should just contain pixels of values 0,1,2,3 if you are working on...

Hi, Thanks for mentioning the 'labelcol' issue. The datasets used in this work have a labelcol folder. So, I think it is better to change the readme. Regarding resizing, for...

Can you please add more description for this error ? For which dataset and which model are you getting this bug ?

I also had the same question. I am also not sure where the 4x4 patch dividing is done. From my understading, the OverlapPatchEmbed is just a convlayer. The img_size argument...

I just picked the first mask i.e _mask

I have used the same parameter settings you have specified. Can you please make sure you are using the model after this [commit ](https://github.com/jeya-maria-jose/UNeXt-pytorch/commit/5106d6b9a8004ef7055051caa8ee5379ab44525c). The previous model file was actually...

Maybe change the random seed and check average over different train-test splits?

Yes, I will make sure to release the pretrained models shortly.