This repository contains all the code to reproduce the experiments made in our paper Domain Siamese CNNs for Sparse Multispectral Disparity Estimation accepted at ICPR 2020.
This is an overview of our proposed architecture.
Here are some examples of the disparity maps with the human silhouettes segmented. The segmentation masks were provided with the dataset.
You can find the dependencies of the project in the requirements.txt file. To install them all, simply type:
pip install -r requirements.txt
Download both datasets in a folder named "litiv" on your computer. For the LITIV 2018 dataset, both the rectified images (used for this paper) and the raw images are available.
This is an example of a possible command to train our network.
python train.py --fold 1 --model domainnet
To see all possible options, simply use:
python train.py --help
Here is an example of you want to test the network:
python test.py --fold 1 \
--model domainnet \
--loadmodel pretrained/domainnet/fold1.pt \
--max_disparity 64 \
--n 3
Once again, to see all available options, use:
python test.py --help
For any comments, questions or concerns, feel free to contact me at david-alexandre.beaupre@polymtl.ca
See the LICENSE file for more details.