No description, website, or topics provided.
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Failed to load latest commit information.


Code for the [Hide-and-Seek: Forcing a Network to be Meticulous for Weakly-supervised Object and Action Localization, ICCV 2017] Krishna Kumar Singh, Yong Jae Lee (

If you use our work, please cite:

  title = {Hide-and-Seek: Forcing a Network to be Meticulous for Weakly-supervised Object and Action Localization},
  author = {Krishna Kumar Singh and Yong Jae Lee},
  booktitle = {International Conference on Computer Vision (ICCV)},
  year = {2017}


  1. Torch (
  2. For training use the code
  3. For the visualization and generating Class Activation Maps(CAM) use the code


  1. Please download train.lua and opts.lua and replace it in
  2. The new code has two additional arguments patchSize and hideProb. patchSize decides the size of the patch to be hidden. For example to hide the patches of size 32 give argument -patchSize 32. Multiple patch sizes can be provided seperated by comma, for example -patchSize 0,16,32,44,56. Here, 0 indicates no patch will be hidden. hideProb indicates by what probability patches will be hidden. For example to hide patches with 50% probability give the argument -hideProb 0.5.
  3. If you need to hide the image patches in MXNet, please refer the code

Pre-trained Models

  1. AlexNet-HaS-Mixed:
  2. GoogLeNet-HaS-32:
  3. Please subtract mean and divide by standard deviation (meanstdCache.t7). For class ordering refer classes.t7.


Please contact for any questions.