dnlnet

Disentangled Non-Local Neural Networks

Introduction

[ALGORITHM]

This example is to reproduce "Disentangled Non-Local Neural Networks" for semantic segmentation. It is still in progress.

Citation

@misc{yin2020disentangled,
    title={Disentangled Non-Local Neural Networks},
    author={Minghao Yin and Zhuliang Yao and Yue Cao and Xiu Li and Zheng Zhang and Stephen Lin and Han Hu},
    year={2020},
    booktitle={ECCV}
}

Results and models (in progress)

Cityscapes

Method	Backbone	Crop Size	Lr schd	Mem (GB)	Inf time (fps)	mIoU	mIoU(ms+flip)	download
dnl	R-50-D8	512x1024	40000	7.3	2.56	78.61	-	model | log
dnl	R-101-D8	512x1024	40000	10.9	1.96	78.31	-	model | log
dnl	R-50-D8	769x769	40000	9.2	1.50	78.44	80.27	model | log
dnl	R-101-D8	769x769	40000	12.6	1.02	76.39	77.77	model | log
dnl	R-50-D8	512x1024	80000	-	-	79.33	-	model | log
dnl	R-101-D8	512x1024	80000	-	-	80.41	-	model | log
dnl	R-50-D8	769x769	80000	-	-	79.36	80.70	model | log
dnl	R-101-D8	769x769	80000	-	-	79.41	80.68	model | log

ADE20K

Method	Backbone	Crop Size	Lr schd	Mem (GB)	Inf time (fps)	mIoU	mIoU(ms+flip)	download
DNL	R-50-D8	512x512	80000	8.8	20.66	41.76	42.99	model | log
DNL	R-101-D8	512x512	80000	12.8	12.54	43.76	44.91	model | log
DNL	R-50-D8	512x512	160000	-	-	41.87	43.01	model | log
DNL	R-101-D8	512x512	160000	-	-	44.25	45.78	model | log