CapsNet-Fashion-MNIST Capsule Network for classification of MNIST Fashion dataset.

Capsule Network is made on the basis of this paper by Geoffrey E. Hinton.

The model is trained on Fashin MNIST dataset. Fashion-MNIST is a dataset of Zalando's article images—consisting of a training set of 60,000 examples and a test set of 10,000 examples. Each example is a 28x28 grayscale image, associated with a label from 10 classes. Zalando intends Fashion-MNIST to serve as a direct drop-in replacement for the original MNIST dataset for benchmarking machine learning algorithms. It shares the same image size and structure of training and testing splits.

Download the dataset here!

Layers used are defined as follows:

====================================================================================================
Layer (type)                     Output Shape          Param #     Connected to                     
====================================================================================================
input_1 (InputLayer)             (None, 28, 28, 1)     0                                            
____________________________________________________________________________________________________
conv1 (Conv2D)                   (None, 20, 20, 256)   20992       input_1[0][0]                    
____________________________________________________________________________________________________
primarycap_conv2d (Conv2D)       (None, 6, 6, 256)     5308672     conv1[0][0]                      
____________________________________________________________________________________________________
primarycap_reshape (Reshape)     (None, 1152, 8)       0           primarycap_conv2d[0][0]          
____________________________________________________________________________________________________
primarycap_squash (Lambda)       (None, 1152, 8)       0           primarycap_reshape[0][0]         
____________________________________________________________________________________________________
digitcaps (CapsuleLayer)         (None, 10, 16)        1486080     primarycap_squash[0][0]          
____________________________________________________________________________________________________
input_2 (InputLayer)             (None, 10)            0                                            
____________________________________________________________________________________________________
mask_1 (Mask)                    (None, 16)            0           digitcaps[0][0]                  
                                                                   input_2[0][0]                    
____________________________________________________________________________________________________
dense_1 (Dense)                  (None, 512)           8704        mask_1[0][0]                     
____________________________________________________________________________________________________
dense_2 (Dense)                  (None, 1024)          525312      dense_1[0][0]                    
____________________________________________________________________________________________________
dense_3 (Dense)                  (None, 784)           803600      dense_2[0][0]                    
____________________________________________________________________________________________________
out_caps (Length)                (None, 10)            0           digitcaps[0][0]                  
____________________________________________________________________________________________________
out_recon (Reshape)              (None, 28, 28, 1)     0           dense_3[0][0]                    
====================================================================================================

Network Adapted from Xifeng Guo.