master
hrnet: Slightly improved weights
Migrate optimizers to torch (#1547)
Extraction Overhaul (#1533)
extract bugfixes: vgg-obstructed model path + unet-dfl output shape
bugfix: Loss function device allocation