search

arnab39 / Semi-supervised-segmentation-cycleGAN

Pytorch implementation of our paper: Revisting Cycle-GAN for semi-supervised segmentation
MIT License
103 stars 18 forks source link

Results after training for 300 epochs not satisfying #21

Open mojoee opened 4 years ago

mojoee commented 4 years ago

After training for 300 epochs with the standard configuration on the cityscapes dataset, except for the crop size, which I had to downsize to 256x256, the result is just a noise of some random pixel values. Does anyone encounter the same issues?

image

XuanangYou commented 3 years ago

Do you know how to solve this problem?

During handling of the above exception, another exception occurred: Traceback (most recent call last): File "F:/SemicycleGAN0/Semi-supervised-segmentation-cycleGAN-master/main.py", line 87, in main() File "F:/SemicycleGAN0/Semi-supervised-segmentation-cycleGAN-master/main.py", line 73, in main model.train(args) File "F:\SemicycleGAN0\Semi-supervised-segmentation-cycleGAN-master\model.py", line 439, in train fake_gt_discriminator = make_one_hot(fake_gt_discriminator, args.dataset, args.gpu_ids) File "F:\SemicycleGAN0\Semi-supervised-segmentation-cycleGAN-master\utils.py", line 347, in make_one_hot onehot = torch.FloatTensor(labels.size(0), C, labels.size(2), labels.size(3)).zero() IndexError: Dimension out of range (expected to be in range of [-3, 2], but got 3)

ruinianxu commented 1 year ago

@mojoee Hi, I am training it on Cityscapes. After around 200 epoches, mIoU is only around 20%. Did you figure out how to address this issue? Many thanks.