Spatial Transformer Networks with IDSIA-like classifier for German Traffic Signs Dataset classification

This repository introduces a TensorFlow-based implementation of STN + IDSIA classifier that is applied to GTSRB (German Traffic Signs Recognition Benchmark) dataset.

This repository provides the source code and the context for my blog post on Medium: https://medium.com/@kirilldaniluk/convnets-series-spatial-transformer-networks-cff47565ae81

The notebook with code: STN_IDSIA_German_Traffic_Signs

STN Transformation Demos:

Grayscale:

Color:

Acknowledgements:

1. Udacity Self-Driving Car Nanodegree for the inspiration and the inital idea for this project.
2. TensorFlow STN Transformer implementation which I use throughout this project.
3. Moodstocks' Torch implementation of STN + IDSIA for the solution and introduction to STNs.
4. Authors of the original paper on spatial transformer paper.