A Keras implementation of CapsNet in the paper:
Sara Sabour, Nicholas Frosst, Geoffrey E Hinton. Dynamic Routing Between Capsules. NIPS 2017
The current average test error = 0.34%
and best test error = 0.30%
.
Differences with the paper:
- We use the learning rate decay with
decay factor = 0.9
andstep = 1 epoch
,
while the paper did not give the detailed parameters (or they didn't use it?). - We only report the test errors after
50 epochs
training.
In the paper, I suppose they trained for1250 epochs
according to Figure A.1? Sounds crazy, maybe I misunderstood. - We use MSE (mean squared error) as the reconstruction loss and
the coefficient for the loss is
lam_recon=0.0005*784=0.392
.
This should be equivalent with using SSE (sum squared error) andlam_recon=0.0005
as in the paper.
Recent updates:
- Add example for multi-gpu training. On two GTX 1080Ti, speed=
55s/epoch
, in contrast to80s/epoch
on a single GTX 1080Ti. - Correct the Routing algorithm. Now the gradients in inner iterations are blocked.
Reorganize the dimensions of Tensors in this part and optimize some operations to speed up.
About
100s / epoch
on a single GTX 1070 GPU. - Rename
dim_vector
todim_capsule
. - Change prior
b
from Variable to constant and move it frombuild
tocall
. Although it is equivalent to the former version, but the current version is easier to understand. Thanks to #15.
TODO
- Conduct experiments on other datasets.
- Explore interesting characteristics of CapsuleNet.
Contacts
- Your contributions to the repo are always welcome.
Open an issue or contact me with E-mail
[email protected]
or WeChatwenlong-guo
.
Step 1. Install Keras>=2.0 with TensorFlow>=1.2 backend.
pip install tensorflow-gpu
pip install keras
Step 2. Clone this repository to local.
git clone https://github.com/XifengGuo/CapsNet-Keras.git
cd CapsNet-Keras
Step 3. Train a CapsNet on MNIST
Training with default settings:
python capsulenet.py
Training with one routing iteration (default 3).
python capsulenet.py --num_routing 1
Other parameters include batch_size, epochs, lam_recon, shift_fraction, save_dir
can be
passed to the function in the same way. Please refer to capsulenet.py
Step 4. Test a pre-trained CapsNet model
Suppose you have trained a model using the above command, then the trained model will be
saved to result/trained_model.h5
. Now just launch the following command to get test results.
$ python capsulenet.py --is_training 0 --weights result/trained_model.h5
It will output the testing accuracy and show the reconstructed images. The testing data is same as the validation data. It will be easy to test on new data, just change the code as you want.
You can also just download a model I trained from https://pan.baidu.com/s/1sldqQo1
Step 5. Train on multi gpus
This requires Keras>=2.0.9
. After updating Keras:
python capsulenet-multi-gpu.py --gpus 2
It will automatically train on multi gpus for 50 epochs and then output the performance on test dataset. But during training, no validation accuracy is reported.
Test Errors
CapsNet classification test error on MNIST. Average and standard deviation results are reported by 3 trials. The results can be reproduced by launching the following commands.
python capsulenet.py --num_routing 1 --lam_recon 0.0 #CapsNet-v1
python capsulenet.py --num_routing 1 --lam_recon 0.392 #CapsNet-v2
python capsulenet.py --num_routing 3 --lam_recon 0.0 #CapsNet-v3
python capsulenet.py --num_routing 3 --lam_recon 0.392 #CapsNet-v4
Method | Routing | Reconstruction | MNIST (%) | Paper |
---|---|---|---|---|
Baseline | -- | -- | -- | 0.39 |
CapsNet-v1 | 1 | no | 0.39 (0.024) | 0.34 (0.032) |
CapsNet-v2 | 1 | yes | 0.36 (0.009) | 0.29 (0.011) |
CapsNet-v3 | 3 | no | 0.40 (0.016) | 0.35 (0.036) |
CapsNet-v4 | 3 | yes | 0.34 (0.016) | 0.25 (0.005) |
Training Speed
About 100s / epoch
on a single GTX 1070 GPU.
About 80s / epoch
on a single GTX 1080Ti GPU.
About 55s / epoch
on two GTX 1080Ti GPU by using capsulenet-multi-gpu.py
.
Reconstruction result
The result of CapsNet-v4 by launching
python capsulenet.py --is_training 0 --weights result/trained_model.h5
Digits at top 5 rows are real images from MNIST and digits at bottom are corresponding reconstructed images.
The model structure:
-
TensorFlow:
- naturomics/CapsNet-Tensorflow
I referred to some functions in this repository. - InnerPeace-Wu/CapsNet-tensorflow
- chrislybaer/capsules-tensorflow
- naturomics/CapsNet-Tensorflow
-
PyTorch:
-
MXNet:
-
Chainer:
-
Matlab: