CapsNet-MXNet

This example is MXNet implementation of CapsNet:
Sara Sabour, Nicholas Frosst, Geoffrey E Hinton. Dynamic Routing Between Capsules. NIPS 2017 - The current best test error is 0.29% and average test error is 0.303% - The average test error on paper is 0.25%

Log files for the error rate are uploaded in repository.

Usage

Install scipy with pip

pip install scipy

Install tensorboard with pip

pip install tensorboard

On Single gpu

python capsulenet.py --devices gpu0

On Multi gpus

python capsulenet.py --devices gpu0,gpu1

Full arguments

python capsulenet.py --batch_size 100 --devices gpu0,gpu1 --num_epoch 100 --lr 0.001 --num_routing 3 --model_prefix capsnet

Prerequisities

MXNet version above (0.11.0)
scipy version above (0.19.0)

Results

Train time takes about 36 seconds for each epoch (batch_size=100, 2 gtx 1080 gpus)

CapsNet classification test error on MNIST

python capsulenet.py --devices gpu0,gpu1 --lr 0.0005 --decay 0.99 --model_prefix lr_0_0005_decay_0_99 --batch_size 100 --num_routing 3 --num_epoch 200

| Trial | Epoch | train err(%) | test err(%) | train loss | test loss | | :---: | :---: | :---: | :---: | :---: | :---: | | 1 | 120 | 0.06 | 0.31 | 0.0056 | 0.0064 | | 2 | 167 | 0.03 | 0.29 | 0.0048 | 0.0058 | | 3 | 182 | 0.04 | 0.31 | 0.0046 | 0.0058 | | average | - | 0.043 | 0.303 | 0.005 | 0.006 |

We achieved the best test error rate=0.29% and average test error=0.303%. It is the best accuracy and fastest training time result among other implementations(Keras, Tensorflow at 2017-11-23). The result on paper is 0.25% (average test error rate).

| Implementation| test err(%) | train time/epoch | GPU Used| | :---: | :---: | :---: |:---: | | MXNet | 0.29 | 36 sec | 2 GTX 1080 | | tensorflow | 0.49 | 10 min | Unknown(4GB Memory) | | Keras | 0.30 | 55 sec | 2 GTX 1080 Ti |