This is step by step implementation of fractionally strided convolutional net Originally written with Tensorflow 0.12 on Python 3.5. If you use different version of python you might see errors or divergence in results If you you use latest version of Tensorflow (1.0 or newer) you might modify few lines of code according to Tensorflow 1.0 API
Note, that you need GPU version of tensorflow, and installed CUDA. Right now tensorflow doesn't support "max_pool_with_argmax" in cpu mode.