Matrix-CapsNet-EM-routing-tensorflow

This is a trial implementation of MATRIX CAPSULES WITH EM ROUTING in TensorFlow framework and Python programming language. （仅供交流学习使用 Education purpose only）

Project completeness 10%. To see a finsihed collaborative repository, see https://github.com/www0wwwjs1/Matrix-Capsules-EM-Tensorflow

Related Work

Link to CapsNet implementation （仅供交流学习使用 Education purpose only）: https://github.com/yhyu13/CapsNet-python-tensorflow

TODO

•  Convert smallNORB dataset into TFRecords

•  Split smallNORB dataset into manageable chunks (optional)

•  Implement Convolutional Capsule Layers

•  Implement EM routing

•  Implement Spread Loss with linear annealing

•  Reproduce test error on smallNORB dataset (if manageable)

•  Reproduce experiments on smallNORB dataset (if manageable)

Dataset

This database is intended for experiments in 3D object reocgnition from shape. It contains images of 50 toys belonging to 5 generic categories: four-legged animals, human figures, airplanes, trucks, and cars. The objects were imaged by two cameras under 6 lighting conditions, 9 elevations (30 to 70 degrees every 5 degrees), and 18 azimuths (0 to 340 every 20 degrees).

The training set is composed of 5 instances of each category (instances 4, 6, 7, 8 and 9), and the test set of the remaining 5 instances (instances 0, 1, 2, 3, and 5).

Download NORB

cd data ./download.sh

The dataset will be available under folder smallNORB.

Write to TFRecord

python dataset.py tfrecord

The tfrecord files will appear under folder data.

Reference:

Dynamic Routing Between Capsules

MATRIX CAPSULES WITH EM ROUTING

THE small NORB DATASET, V1.0