CapsNet-PT
A PyTorch implementation of CapsNet based on Geoffrey Hinton's paper Dynamic Routing Between Capsules Capsule is a magic model, before implement it, i saw lots of pytorch implementation, but lots of them suffer from different mistakes, such as wrong softmax dim, wrong squash dim. However, all these models give acceptable results.:) Although all the models can be used in practice, Hinton's capsule model originates from Cognitive Neuroscience ,we should judge this model from a biological perpective instead of just checking its accuracy.
There is another very good youtube video for capsule explanation Capsule Networks, you can view it for better understanding.
Steps
Step 1.
Clone this repository with git.
$ git clone https://github.com/Primus-zhao/CapsNet-PT.git
$ cd CapsNet-PT
Step 2.
Download the MNIST dataset, mv and extract it into data/raw directory, then run python gen_pt.py to generate pt data for pytorch
$ wget -c -P data/raw http://yann.lecun.com/exdb/mnist/{train-images-idx3-ubyte.gz,train-labels-idx1-ubyte.gz,t10k-images-idx3-ubyte.gz,t10k-labels-idx1-ubyte.gz}
you can also download the data from lecun's homepage and put it in data folder, it's faster
$ gunzip data/raw/*.gz
the last step is for converting original binary file to pt file for pytorch
$ python gen_pt.py
Step 3. Start the training:
$ python main.py
The default parameters of batch size is 128, and epoch is 3, we will validate after 100 steps. You can also configure this with display_step variable in main.py. Be careful, since validation will go through 10000 test images, validation will be a bit slow, so don't set display_step too small