Tarun K

Yes, that is a limitation in my code currently.

@jiyangbai Hey, did you manage to find the answer to this? Looks like it takes really long time, any idea how much time roughly it takes to run one experiment?

I am facing the same trouble. The adversarial loss is diverging and loss_D is going to 0 very quickly. could you mention the hyper-parameter setting to reproduce the feature space...

loss_adv is the loss that a target sample is classified as the source, so the segmentation network should minimize this loss to correctly confuse the discriminator (discriminator is already learning...

I am aware of the loss used on MCD paper, it is given as follows. https://github.com/mil-tokyo/MCD_DA/blob/af10217c5c5451dcd8bc3e975a7d067c285cc029/visda_classification/res_train_main.py#L145 which seems to be different to the loss you used . https://github.com/lijin118/CGDM/blob/2d8fee0c8df158fb078ceed65979abbe46282506/utils.py#L419 Can you...

The loss doesn't make sense to me. In the paper, you state that the difference is computed between (feature norm computed using last parameters) and (radius computed using current parameters)....

Looks like it has to do something with the versions of torchvision.io. Can you try to match the versions recommended in the repo and let me know if the problem...

Hi, Sorry for the late reply. We did not encounter this problem. Can you try using some other library for writing out the frames to file?

Hi, can you provide with the pre-trained models?