ContourNet

Richer Convolutional Features for Contour Detection on PASCAL VOC 2012 Dataset.

Introduction

My ContourNet(CNet) can be used to detection the object contour on PASCAL VOC 2012 Dataset. CNet refer RCF, DFN and CEDN.

Dataset and Preprocessed

In CEDN, they proposed the method of preprocessing data. Actually, they have provided the preprocessed-data.

My Results

Method	Original Motivation	ODS F-score	OIS F-score	AP
CNet(Ours)	Contour Detection	0.494	0.509	0.373
CEDN	Contour Detection	0.486	0.500	0.354
RCF	Edge Detection	0.459	0.475	0.333
HED	Edge Detection	0.441	0.454	0.311

note: The same environment means the same data augmentation, the same framework, the same time seed and so on.

Environment

pytorch-1.0.0, numpy, PIL

Thanks

Thanks to XuanYi Li, Wei Shen.

Contact

zhangchbin AT gmail.com