Side-output Residual Network for Object Symmetry Detection in the Wild

SRN is build on Holistically-Nested Edge Detection (HED) [1] with Residual Unit (RU). RU is used to compute the residual between output image and side-output of SRN.

1. Install prerequisites for Caffe (http://caffe.berkeleyvision.org/installation.html#prequequisites)
2. Build HED (https://github.com/s9xie/hed)
3. Copy folder SRN to 'example/' in HED.

1. Download benchmark Sym-PASCAL trainning set Our dataset Sym-PASCAL derived from PASCAL 2011 segmentation dataset [1]. Download：(OneDrive) or (BaiduYun)
2. Download the Pre-trained VGG [3] model. (VGG19) Copy it to 'example/SRN/'
3. Change the dataset directory of 'train_val.prototxt'
4. Change the directory of solver.prototxt
5. Run solve.py

1. Change the dataset directory of 'SRNtest.py'
2. run 'SRNtest.py'

We use the evaluation code of [3] to draw the PR curve. The code can be download (spb-mil)

(Pre-trained SRN model on Sym-PASCAL)

(Sym-PASCAL)

(SYMMAX)

(WH-SYMMAX) mostly taken from http://wei-shen.weebly.com/publications.html

(SK506) mostly taken from http://wei-shen.weebly.com/publications.html

Ref

[1] S. Xie and Z. Tu. Holistically-nested edge detection. In International Conference on Computer Vision, 2015

[2] M. Everingham, L. Van Gool, C. K. I. Williams, J. Winn, and A. Zisserman. The PASCAL Visual Object Classes Challenge 2011 (VOC2011) Results. http://www.pascal-network.org/challenges/VOC/voc2011/workshop/index.html.

[3] S. Tsogkas and I. Kokkinos. Learning-based symmetry detection in natural images. In European Conference on Computer Vision