Skip to content

Latest commit

 

History

History
20 lines (14 loc) · 751 Bytes

README.md

File metadata and controls

20 lines (14 loc) · 751 Bytes

Capsule Networks

Capsule Networks are networks that are considered to perform "inverse graphics". This means that using the image in the view they give us the embeddings which can be used to reconstruct the image perfectly well.

In this repository, I created a capsule network inspired by this paper. The network is built entirely with tensorflow's estimator API.

Architecture

The architecture of the network is as shown below.

capsule nets

Reconstruction Results

results

Graph As Shown on TensorBoard

tensorboard graph

Dependencies

  1. TensorFlow
  2. Matplotlib
  3. Numpy