Mashiro

It seems that your pip connection error, can you download other packages via pip? Maybe you can try to close your proxy.

You can install mmcv with our pre-built package, such as `pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/cu111/torch1.8.0/index.html`. The PyTorch version and Cuda version should be consistent with your conda enviroment.

> This is the initial version of implementation of VFL and I forgot to refine it. > `alpha * (pred_sigmoid - target).abs().pow(gamma) * (target

Ohhh! Thanks, I understand it now.

You can just change '--crop_size' in train.py... I can train the net with '--bs_mult=3', '--crop_size=336', '--bs_mult_val=1'（GPU memory 8G）

Actually， this project only supports wideresnet because there is no code about(resnet50 or resnet101) in gscnn.py(Although the net is defined in file network).

make --syncbn=False in train.py. Besides, I also delete some data augmention and change its image transformation because the loss is hard to converge for single GPU card based on released...

@shifangtian You can follow my repo https://github.com/HAOCHENYE/GSCNN-for-Single-GPU This net is very hard to training on single GPU,my mean iou only reached 0.4084

https://github.com/nv-tlabs/GSCNN/issues/34#issuecomment-564370469 You can try it

Hi! I'm training the model with COCO now. Can you share me you training settings?My loss seems to be higher than MOT17. BTW, my bbox's score is very low too...