Home

Awesome

Group Whitening

This project provides the PyTorch implementation of Group Whitening described in the following paper:

Group Whitening: Balancing Learning Efficiency and Representational Capacity,

Lei Huang, Yi Zhou, Li Liu, Fan Zhu, Ling Shao

IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2021. arXiv:2009.13333.

Highlights

Table 1, ImageNet Classification. Standard setup: batchSize=256, wd=0.0001, init lr=0.1, 100 epochs (30,60,90- decay).

BaseLineUsing GW module
ResNet-5076.2377.72 (model, pth )
ResNet-10177.6978.71 (model, pth )
ResNeXt-5077.0178.43 (model, pth )
ResNeXt-10179.2980.43 (model, pth )

Table 2, COCO object detection results using Faster R-CNN with ResNet-50+FPN. We use the 1x lr scheduling (90k iterations), with a batch size of 16 on 8 GPUs.

2fc head box4conv 1fc head box
BN-frozen36.31% AP36.39 % AP
GN36.32% AP37.86 % AP
GW38.13% AP39.60 % AP

Table 3, COCO object detection and segmentation results using Mask R-CNN with ResNeXt-101+FPN. We use the 1x lr scheduling (180k iterations), with a batch size of 8 on 8 GPUs.

AP (box)AP(mask)
BN-frozen42.24%37.53%
GN42.18%37.54%
GW44.41%39.17%

Contact

huanglei36060520@gmail.com

Acknowledgement

Note that the code repo ./classification/ is based on the IterNorm project, and the code repo /maskrcnn_debug/is based on the maskrcnn project.