Source code for the paper "Three for one and one for three: Flow, Segmentation, and Surface Normals" (BMVC 2018)
If you find the repository useful, please consider citing the paper
@inproceedings{le18bmvc,
author = {Hoang-An Le and Anil Baslamisli and Thomas Mensink and Theo Gevers},
title = {Three for one and one for three: Flow, Segmentation, and Surface Normals},
booktitle = {British Machine Vision Conference (BMVC)},
year = {2018},
}
Parts of the Virtual KITTI dataset are included in this repository for your convenience. If used, please cite the corresponding paper following the instructions at NAVER LABS EUROPE
Ground truth surface normals are converted from the provided depth images using the method given by Barron and Malik, Shape, Illumination, and Reflectance from Shading. The source code is available in the project page or [here](). Please cite the according paper if you use the codes in your research.
Build caffe
Each run is named in the following format
[data]-[target]-[input],
where
[data]
is either gdextracted
for the Nature dataset or vkitti
for the
Virtual KITTI dataset.[target]
is the target modality, written with rf_
prefix, e.g. rf_flow
indicates the experiment to refine optical flow modality.[input]
is the input branch configuration
gt
indicates ground truth modalities and pr
the predicteds.pr
_
indicate that the modalities are concatenated along the
depth channels-
indicate that the modalities are separated in different input
branchesflow
for optical flowseg
for semantic segmentationnorm
for surface normalYou can either call caffe train command directly using the prototxts in the runs directory or use our python wrapper