unet

U-Net: Convolutional Networks for Biomedical Image Segmentation

Reference

Ronneberger O, Fischer P, Brox T. U-net: Convolutional networks for biomedical image segmentation[C]//International Conference on Medical image computing and computer-assisted intervention. Springer, Cham, 2015: 234-241.

Performance

Cityscapes

Model	Backbone	Resolution	Training Iters	Batch Size	mIoU	mIoU (flip)	mIoU (ms+flip)	Links
UNet	-	1024x512	160000	4	65.00%	66.02%	67.29%	model | log | vdl

STARE

Model	Backbone	Resolution	Training Iters	Batch Size	AUC ROC	DICE	mIoU	Links
UNet	-	128x128	40000	16	95.93%	90.43%	83.54%	model | log | vdl

DRIVE

Model	Backbone	Resolution	Training Iters	Batch Size	AUC ROC	DICE	mIoU	Links
UNet	-	128x128	40000	16	95.58%	89.50%	82.12%	model | log | vdl

CHASE DB1

Model	Backbone	Resolution	Training Iters	Batch Size	AUC ROC	DICE	mIoU	Links
UNet	-	128x128	40000	16	95.69%	88.54%	80.87%	model | log | vdl

HRF

Model	Backbone	Resolution	Training Iters	Batch Size	AUC ROC	DICE	mIoU	Links
UNet	-	256x256	40000	16	93.39%	86.83%	78.45%	model | log | vdl