Grigoriy Livshits
Grigoriy Livshits
I also need SVD. My particular choice is PyTorch, but I've mentioned Tensorflow just to keep some sort of "unity" with other DL frameworks :)
I have found the same for me: training crashes at 5024 iterations. This is caused by automatic enabling of additional loss, namely path regularization (PL). So you should choose the...
Hello. I don't know about the FFHQ, but on my domain, DCL worked better than Hinge loss.
Same. Non-fair memory distribution in multi-gpu setting.
I think the problem is in the following code: pl_noise = torch.randn(images.shape, device=device) / math.sqrt(num_pixels) The noise becomes batchsize-dependent, thus occupying more space on GPU when retaining the graph.
Some sort of "bubble-checkboard".
I think that it is related to previous issue with jinc. Without 2fc this function falls very aggressive, cutting out most details in a vicinity of a point on featuremap.
I've used 0.1 lr multiplier for mapping. Note that this is not valid for all domains. But nevertheless, the more I disable noise - the more uniform images are. My...
I think I've found an issue. In your code (and in original code of Nvidia) the blur kernel [1,3,3,1] is used. First of all, I don't see any reason for...
Replaced PIL.Image.open with torchvision.io.read_image (it outputs uint8 tensor), still leaks. ``` import torch import torchvision from torchvision import transforms class TestDataset(torch.utils.data.Dataset): def __init__(self, images): self.images = images self.init_base_transform() def __len__(self):...