kaiyuyue/cgnl-network.pytorch

CGNL for Image Segmentation

usb1508 opened this issue · 2 comments

How do we use CGNL in image segmentation tasks?

Hello,

Whatever frameworks are used there, the backbone is the target place to add CGNL blocks.

For example, in Detectron2, assuming the backbone is ResNet, the CGNL blocks should be added into detectron2/modeling/backbone/resnet.py. The modification way is the same way for image classification.

Then FPN for instance segmentation or Panoptic FPN for semantic segmentation can be built upon the CGNL-ResNet. Just like the typical way to build the meta arch for detection, segmentation, or other tasks you wanna do.

Here is the list you can refer to: Detectron2 Backbones and MMDetection Backbones.
Hope this could be helpful.

@kaiyuyue Thanks for the details.