SCNet Code

Region matching code is contributed by Kai Han ([email protected]).

Dense matching code is contributed by Rafael S. Rezende ([email protected]).

This code is written in MATLAB, and implements the SCNet[1]. For the dataset, see our project page: http://www.di.ens.fr/willow/research/scnet.

Install Dependencies

• Install [MatConvNet] (http://www.vlfeat.org/matconvnet/).
• Download [VLFeat] (http://www.vlfeat.org/) [0.9.20] in './utils/'
• Download the following source codes of object proposal or other proposal methods you would like to test:
  • [Randomized Prim’s] (https://github.com/smanenfr/rp#rp) (our preference);
  • [EdgeBox] (https://github.com/pdollar/edges);
  • [SelectiveSearch] (http://koen.me/research/selectivesearch/);
  • [Multiscale Combinatorial Grouping] (https://github.com/jponttuset/mcg);
• Download a ImageNet [Caffe Reference model] (http://www.vlfeat.org/matconvnet/pretrained/) in ./data/models/.

Codes

SCNet_Matconvnet

Additional Matconvnet modules implemented for SCNet. These code should be copied into matconvnet/matlab/ folder.

SCNet

This is the primary net work training and testing code.

• SCNet_A_init.m, SCNet_AG_init.m, SCNet_AGplus_init.m: initialize the SCNet_A, SCNet_AG, SCNet_AG+.

• SCNet_A.m, SCNet_AG.m, SCNet_AGplus.m: train SCNet_A, SCNet_AG, SCNet_AG+.

• eva_PCR_mIoU_SCNet_A.m, eva_PCR_mIoU_SCNet_AG.m, eva_PCR_mIoU_SCNet_AGplus.m: evaluate the trained nets.

• eva_PCR_mIoU_ImageNet_SCNet_A.m, eva_PCR_mIoU_ImageNet_SCNet_AG.m, eva_PCR_mIoU_ImageNet_SCNet_AGplus.m: evaluate SCNets with ImageNet pretrained parameters, i.e., SCNets without training.

SCNet_Baselines

Comparison code for our SCNet features and HOG features with NAM, PHM and LOM in Proposal Flow [2, 3].

• NAM_HOG_eva.m, PHM_HOG_eva.m, LOM_HOG_eva.m: evaluate NAM, PHM, and LOM with HOG features.

• NAM_SCNet_eva.m, PHM_SCNet_eva.m, LOM_SCNet_eva.m: evaluate NAM, PHM, and LOM with learned SCNet features.

• HOG_SCNet_AG_eva.m: replace the learned SCNet feature by HOG feature in SCNet_AG model.

Data

We used PF-PASCAL, PF-WILLOW, PASCAL Parts and CUB data sets and follows Proposal Flow[2, 3] to generate our trainging data.

Triaining data preparation code is put in PF-PASCAL-code folder.

Notes

• The code is provided for academic use only. Use of the code in any commercial or industrial related activities is prohibited.
• If you use our code or dataset, please cite the paper.

@InProceedings{han2017scnet,
author = {Kai Han and Rafael S. Rezende and Bumsub Ham and Kwan-Yee K. Wong and Minsu Cho and Cordelia Schmid and Jean Ponce},
title = {SCNet: Learning Semantic Correspondence},
booktitle = {International Conference on Computer Vision (ICCV)},
year = {2017}
}

References

[1] Kai Han, Rafael S. Rezende, Bumsub Ham, Kwan-Yee K. Wong, Minsu Cho, Cordelia Schmid, Jean Ponce, "SCNet: Learning Semantic Correspondence", International Conference on Computer Vision (ICCV), 2017.

[2] Bumsub Ham, Minsu Cho, Cordelia Schmid, Jean Ponce, "Proposal Flow: Semantic Correspondences from Object Proposals", IEEE Trans. on Pattern Analysis and Machine Intelligence (TPAMI), 2017

[3] Bumsub Ham, Minsu Cho, Cordelia Schmid, Jean Ponce, "Proposal Flow", Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016