Denis Kuznedelev

@janhenrikbern yeah, I've faced the same issue, when trying running it on colab with 16gb gpu. Even reduction to `image_size=64, dim=128` doesn't help, and even it seems that it changes...

@janhenrikbern try to decrease the size of batch, `default_batch_size = 8` would not fit the `16gb` GPU, I've decreased batch size to `2` and now I occupy `9.5gb` of memory.

@janhenrikbern unfortunately after 10000 iteration training terminated again, one needs to keep in mind, that the resolution is increased gradually from 32 to 128, therefore the batches become more heavy....

@janhenrikbern any attempt to run a `pi-GAN` with the image resolution up to 128 fails even with the smallest size of `batch=1`. If one restricts himself to small final resolutions...

@krips89 we have created our own version on `PyTorch Lightning` an it run succesfully, however, with our computing resources the obtained quality was poor

@janhenrikbern @krips89 actually, I've found a problem in this implementation. When performing accumulation of the generator and discriminator loss, grads of the loss are not detached - hence, stored in...

@janhenrikbern The adapted version in PyTorch lighting implemented by me works - but the results are not satisfactory. Actually, as mentioned in other issues, there is some discrepancy between the...

Hi, @thephet , I am on Ubuntu 20.04 LTS, the python version is 3.8.2, tensorflow.__version__ - '2.2.0', keras.__version__ - '2.4.2'. Initially, I was running code on the master branch, however,...

@ShaashvatShetty you should have the following structure in your xView directory: ``` # parent # xView # └── train_images # └── val_images (may be empty directory) # └── xView_train.geojson ```

@QuentinAndre11 300 epochs may not suffice, since the dataset is quite small in terms of number of images, but the images themselves contain many instances. I think, one should train...