naturomics/CapsNet-Tensorflow
A TensorFlow implementation of Geoffrey Hinton's CapsNet architecture, a neural network designed to replace traditional CNNs with capsule-based routing for image classification.
Velocity · 7d
+1.2
★ / day
Trend
→steady
star history
This repository provides a complete TensorFlow implementation of CapsNet, based on Hinton’s paper on dynamic routing between capsules. It achieves 99.64% accuracy on MNIST and 90.60% on Fashion-MNIST. The implementation includes the core capsule routing algorithm, primary capsule layers, and digit capsule layers. A multi-GPU distributed version is also available.