We have designed and developed an interactive system that allows users to experiment with deep learning image classifiers and explore their robustness and sensitivity. Selected areas of an image can be removed in real time with classical computer vision inpainting algorithms, allowing users to ask a variety of "what if" questions by experimentally modifying images and seeing how the deep learning model reacts. The system also computes class activation maps for any selected class, which highlight the important semantic regions of an image the model uses for classification. The system runs fully in browser using Tensorflow.js, React, and SqueezeNet. An advanced inpainting version is also available using a server running the PatchMatch algorithm from the GIMP Resynthesizer plugin.
This is the code repository for the accepted CVPR 2018 Demo: Interactive Classification for Deep Learning Interpretation. Visit our research group homepage Polo Club of Data Science at Georgia Tech for more related research!
The modified image (left), originally classified as dock is misclassified as ocean liner when the masts of a couple boats are removed from the original image (right). The top five classification scores are tabulated underneath each image.
Download or clone this repository:
git clone https://github.com/poloclub/interactive-classification.git
Within the cloned repo, install the required packages with yarn:
yarn
To run, type:
yarn start
The following steps are needed to set up PatchMatch inpainting, which currently only works on Linux:
- Clone the Resynthesizer repository and follow the instructions for building the project (stop after running
make
) - Find the
libresynthesizer.a
shared library in the generatedlib
folder and copy it to theinpaint
folder in this repository - Run
gcc resynth.c -L. -lresynthesizer -lm -lglib-2.0 -o prog
(may have to install glib2.0 first) to generate theprog
executable - You can now run
python3 inpaint_server.py
and PatchMatch will be used as the inpainting algorithm when running the React application withyarn start
.
Interactive Classification for Deep Learning Interpretation
Angel Cabrera, Fred Hohman, Jason Lin, Duen Horng (Polo) Chau
Demo, Conference on Computer Vision and Pattern Recognition (CVPR). June 18, 2018. Salt Lake City, USA.
@article{cabrera2018interactive,
title={Interactive Classification for Deep Learning Interpretation},
author={Cabrera, Angel and Hohman, Fred and Lin, Jason and Chau, Duen Horng},
journal={Demo, Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
year={2018},
organization={IEEE}
}
Name | Affiliation |
---|---|
Angel Cabrera | Georgia Tech |
Fred Hohman | Georgia Tech |
Jason Lin | Georgia Tech |
Duen Horng (Polo) Chau | Georgia Tech |
MIT License. See LICENSE.md
.
For questions or support open an issue.