Efficient mean-field inference and learning in dense CRFs. This repository contains both CPU and GPU implementations of high-dimensional filtering ops using permutohedral lattice.
First, build the op with cd filter_ops ; make .
Make sure CUDA_HOME is set as well as appropriate flags in filter_ops/Makefile.
Then, see the notebook op_test.ipynb for an example.
Tested on:
- Python 3.6
- TensorFlow 1.4.1
- Titan X Pascal