John Sutor
John Sutor
I can implement this.
@deepsthewarrior yep!
Just adding on for visibility, this one is pretty solid too: https://github.com/postlight/parser
I had a similar turnout after 500,000 steps, showing very little to no improvement from the 150000 step example.
I can take this issue.
I'll investigate and get back to you!
Seems fairly straightforward to achieve based on https://github.com/keyu-tian/SparK/tree/main/pretrain#regarding-sparse-convolution. I don't mind giving it a stab, my thoughts are to implement the encoder and decoder from their code base (https://github.com/keyu-tian/SparK/tree/main/pretrain) within...
I went ahead and implemented a resnet compatible with the standard `torchvision` library, so that we don't have to add `timm` as a dependency. Furthermore, I achieved passing the mask...
Hey, thanks for checking it out! In regards to your bullets: 1. I think that approach of storing the sparse mask on the modules works (this was going to be...
The feature map channels are used in step three of the forward process, where the hierarchical dense features are calculated for decoding. When the SparK module is created, it creates...