Unsupervised Adversarial Depth Estimation using Cycled Generative Networks

The code for "Unsupervised Adversarial Depth Estimation using Cycled Generative Networks" in 3DV2018
Paper link: https://arxiv.org/pdf/1807.10915.pdf
By Andrea Pilzer, Dan Xu, Mihai Puscas, Elisa Ricci, Nicu Sebe

Content

The experiments are performed on a HPC server with Python 3.6 and Tensorflow 1.10.

1. Training and testing

Training

python main.py --dataset kitti --filenames_file utils/filenames/eigen_train_files_png.txt \
--data_path /path/to/KITTI/ --do_stereo --train_branch b2a

Testing

python main.py --mode test --dataset kitti --filenames_file utils/filenames/eigen_test_files_png.txt \
--data_path /path/to/KITTI/ --do_stereo --checkpoint_path my_model/model-5000

Please note that there is NO extension after the checkpoint name

Evaluation

python utils/evaluate_kitti.py --split kitti --predicted_disp_path my_model/disparities.npy \
--gt_path ~/data/KITTI/

2. Datasets

We used the KITTI dataset in our experiments. Please refer to a very well written dataset description section of Monodepth for data preparation.

3. Trained model

The pretrained model can be downloaded from Google Drive.

4. Citation

Please condiser citing our paper if you find the code is useful for your projects:

@inproceedings{pilzer2018unsupervised,
  title={Unsupervised Adversarial Depth Estimation using Cycled Generative Networks},
  author={Pilzer, Andrea and Xu, Dan and Puscas, Mihai and Ricci, Elisa and Sebe, Nicu},
  booktitle={2018 International Conference on 3D Vision (3DV)},
  pages={587--595},
  year={2018},
  organization={IEEE}
}

@article{pilzer2019progressive,
  title={Progressive Fusion for Unsupervised Binocular Depth Estimation using Cycled Networks},
  author={Pilzer, Andrea and Lathuili{\`e}re, St{\'e}phane and Xu, Dan and Puscas, Mihai Marian and Ricci, Elisa and Sebe, Nicu},
  journal={IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI)},
  year={2019},
  publisher={IEEE}
}