This code is based on Philipp Krähenbühl Efficient Inference in Fully Connected CRFs with Gaussian Edge Potentials
.
http://graphics.stanford.edu/projects/drf/ If you're using this code in a publication, please also cite their papers.
This function is called by grid_sensor for 3D CRF. We also provide a 2D CRF example using data under grid_sensor/data_kitti
rosrun dense_crf comare_hier_dense_crf
It takes around 30s to optimize the full 2D KITTI image. Results also saved under test_data/. Sometimes hierarchical and dense CRF might have small difference.
-
Add dense pairwise potential with high order cliques, which can be used for general 2D or 3D CRF optimization.
-
Adapted for CRF 3D grid optimization and superpixel utils.
-
Change to ros catkin.