/mnist-inpainting

Yet another repository of experiments with RNNs in TensorFlow using MNIST

Primary LanguagePython

MNIST Inpainting

Adam Hornsby

This repository contains all code, model files, visualisations and reports for my experiments with RNNs in TensorFlow. I used recurrent neural networks to classify and predict the next pixel of MNIST images.

alt text

Folder structure

This repository contains the following folders:

  1. code - This contains all .py files necessary for training, evaluating and loading models
  2. visualisations - This contains animated GIFs of the MNIST images being completed by 4 different Gated Recurrent Unit models (GRUs)

Model development files

Files used to train models, generate visualisations and evaluate models are available in the code folder (except the load_models.py file). The run_lstm.py file is the main code used to train both sets of RNN models. The run_generator.py file is the main code used to in-paint images described in section 2 and 3 of this coursework.