Batch Normalization with Enhanced Linear Transformation

BNET has been accepted by T-PAMI. This repository provides the PyTorch implementation of Batch Normalization with Enhanced Linear Transformation

Update

• 11.27.2020: Pretrained models are released.
• 11.26.2020: The code of Image Classification and Object Detection are released.

Models

• Classification:[Google Drive]

• Detection:[Google Drive]

Model Zoo

Backbone	BN	Lr schd	Inf time (fps)	box AP	Download
Faster-R-50-FPN	BN	1x	18.0	37.5	model
Faster-R-50-FPN	BNET-3	1x	17.1	39.5	model
Faster-R-50-FPN	BNET-5	1x	16.3	40.1	model
Faster-R-50-FPN	BNET-5	2x	16.3	42.4	model
Faster-R-50-FPN	BNET-7	1x	15.5	40.7	model
Faster-R-101-FPN	BN	1x	13.4	39.4	model
Faster-R-101-FPN	BNET-3	1x	13.0	40.7	model
Faster-R-101-FPN	BNET-5	1x	12.2	41.8	model
Faster-R-101-FPN	BNET-5	2x	12.2	43.7	model

Backbone	BN	Lr schd	Inf time (fps)	box AP	Download
Retina-R-101-FPN	BN	1x	12.8	38.5	model
Retina-R-101-FPN	BNET-5	1x	11.4	40.7	model

Backbone	BN	Lr schd	box AP	mask AP	Download
Mask-R-50-FPN	BN	1x	38.2	34.7	model
Mask-R-50-FPN	BNET-3	1x	40.2	36.4	model
Mask-R-50-FPN	BNET-5	1x	40.8	36.7	model
Mask-R-101-FPN	BN	1x	40.0	36.1	model
Mask-R-101-FPN	BNET-3	1x	42.2	37.8	model
Mask-R-101-FPN	BNET-5	1x	42.5	37.9	model

Citation

If you use BNET in your research, please cite this project.

@article{BNET,
  title   = {Batch Normalization with Enhanced Linear Normalization},
  author  = {Xu, Yuhui and Xie, Lingxi and Xie, Cihang and Mei, Jieru and
             Qiao, Siyuan and Wei, Shen and Xiong, Hongkai and Alan, Yuille},
  journal= {arXiv preprint arXiv:2011.14150},
  year={2020}
}