README.md

AutoDispNet

Code accompanying the paper: AutoDispNet: Improving Disparity Estimation with AutoML (ICCV 2019). Parts of this codebase is inspired from DARTS.

Note: We provide deployment code only.

Setup

• tf 1.12
• Compile and install lmbspecialops.
• Install netdef_slim.
• Clone this repository.

Running networks

• Change your directory to the network directory (autodispnet/nets).

• Download pre-trained weights with download_weights.sh. Pre-trained weights are provided for networks trained on FlyingThings (CSS, css) and fine-tuned on KITTI (CSS-KITTI). css is a network with smaller memory footprint (see paper for details).

• Go to a network directory (Eg: autodispnet/nets/CSS) and use the following command to test the network on an image pair:

  python3 controller.py eval image0_path image1_path out_dir

• The output is stored in a binary format with .float3 extension (Information on reading the output is here).

Citation

If you use the code or parts of it in your research, you should cite the aforementioned paper:

@InProceedings{SMB19,
  author       = "T. Saikia and Y. Marrakchi and A. Zela and F. Hutter and T. Brox",
  title        = "AutoDispNet: Improving Disparity Estimation With AutoML",
  booktitle    = "IEEE International Conference on Computer Vision (ICCV)",
  month        = "October",
  year         = "2019",
  url          = "http://lmb.informatik.uni-freiburg.de/Publications/2019/SMB19"
}

Author

Tonmoy Saikia ([email protected])