Finding Distractors In Images (CVPR 2015)
Matlab Mathematica
Switch branches/tags
Nothing to show
Clone or download
ohadf Update
Fixed broken link
Latest commit a6aba66 Mar 14, 2017
Failed to load latest commit information.
calculate-features first commit Jul 13, 2015
distractor-prediction first commit Jul 13, 2015
external Update Mar 14, 2017
train-new-model first commit Jul 13, 2015
.gitignore first commit Jul 13, 2015
Lenna.png first commit Jul 13, 2015 Update Jan 29, 2016
demo_predict.m first commit Jul 13, 2015
demo_train_model.m first commit Jul 13, 2015
install_all.m first commit Jul 13, 2015

Finding Distractors In Images

This file contains code for our CVPR 2015 paper. If you use the code, please add the following cite:

  author = {Fried, O. and Shechtman, E. and Goldman, D.B. and Finkelstein, A.}, 
  booktitle = {Computer Vision and Pattern Recognition (CVPR), 2015 IEEE Conference on},
  title = {Finding distractors in images},
  year = {2015},
  pages = {1703-1712},
  doi = {10.1109/CVPR.2015.7298779},
  month = {jun}


The code has many external dependancies. First and foremost, an early version of the code was based on "learning to predict where humans look", for which we are very grateful. Since parts of that code are still present, please consider citing appropriately.

You will need to install other external packages. Please refer to ./external/ for installation instructions.

Getting started

After completing the installation process in ./external/, the demo files should execute properly. We supply two demos:

  • demo_predict.m calculates a distractor prediction map for a given image, using our precalculated model. Start here if you want to predict distractor locations.
  • demo_train_model.m trains a new distractor prediction model. Start here if you want to augment our model with new features and algorithms.


The code was tested on Mac OS X 10.10, using Matlab R2015a. It should work on other configurations with minor adjustments. If you happen to make such adjustments, it would be great if you are willing to share! Propose a change via GitHub and I'll integrate your changes for everyone to enjoy.

For any questions, comments or suggestions, please contact Ohad Fried

Version 1.0