Some code (TensorFlow) based on the paper:
A Kendall, Y Gal, “What Uncertainties Do We Need in Bayesian Deep Learning for Computer Vision?”, NIPS 2017 arXiv
DISCLAIMER: This is NOT the official repo. It is just based on my understanding of the paper. Any feedback is welcome.
I am training a simple autoencoder (regression) to reconstruct MNIST digits.
Download MNIST:
./download.sh
Rescale and save in a python dictionary (possibly resize):
python prepro.py
This repository is released under the MIT LICENSE.