tlkvstepan / mc-cnn-ws

Weakly supervised learning of deep metrics for stereo reconstruction, ICCV17

Weakly supervised learning of deep metrics for stereo reconstruction (MC-CNN-WS)

Please cite our paper if you use code from this repository in your work.

@inproceedings{tulyakov2017weakly,
  title={Weakly supervised learning of deep metrics for stereo reconstruction},
  author={Tulyakov, Stepan and Ivanov, Anton and Fleuret, Francois},
  booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
  pages={1339--1348},
  year={2017}
}

This code requires mc-cnn.

The training can be performed by running:

TRAIN_NET.lua <mode> <method> <arch> <set>

<mode>   - normal / debug.
<method> - mil / contrastive / mil-contrastive / contrastive-dp
<arch>   - fst-kitti / fst-mb
<set>    - kitti / kitti_ext / kitti15 / kitti15_ext / mb