Chainer implementation of CapsNet for MNIST.
For the detail, see Dynamic Routing Between Capsules, Sara Sabour, Nicholas Frosst, Geoffrey E Hinton, NIPS 2017.
python -u train.py -g 0 --save saved_model --reconstruct
Test accuracy of a trained model (without reconstruction) reached 99.66%. For making training easy, this code replaced relu with leaky relu with a very small slope (0.05).
These scripts require the following python libraries.
- Chainer 2.0.0+
- numpy
- matplotlib
For analyzing a trained model, you can produce images reconstructed from digit capsules.
python visualize.py -g 0 --load saved_model
The top green images are real images which are given to the model. Blue images in i-th represents reconstructed ones of digit "i".
If an correct digit is selected as a target, the model reconstructs an image well (see the diagonal cells).
If an irrelevant target is selected, the reconstructed image gets spoiled (see "0" and the others in the column leftmost), maybe because of lack of information in its digit capsule. However, reconstruction toward a relevant target is not always spoiled, even if a target is not correct (see "8" and "9" the column rightmost).
Here, we show reconstructed images after linearly tweaking the value in a dimension in the capsule (as well as section 5.1 and figure 4 in the paper). Green images in the center are reconstructed images without perturbation. Note that a dimension has a different factor if the digit capsule differs, because each matrix for reconstructing each digit is unshared.
You can find and enjoy some factors of variation.
@incollection{NIPS2017_6975,
title = {Dynamic Routing Between Capsules},
author = {Sabour, Sara and Frosst, Nicholas and Hinton, Geoffrey E},
booktitle = {Advances in Neural Information Processing Systems 30},
year = {2017},
}
MIT License. Please see the LICENSE file for details.