ThisLooksLikeThat

paper

This looks like that: deep learning for interpretable image recognition

by Chaofan Chen, Oscar Li, Alina Barnett, Jonathan Su, Cynthia Rudin

usage

demo_cifar2.py

input train.lst/test.lst, output trained net and proto.pkl

show.prototype.py

parse proto.pkl and save prototype as image

CMakeLists.txt

some codes implemented by cpu codes to save

test

• results/cifar-29400.params
  batch size = 20 train for 29400 interation, test accuracy 69%
• results prototxt