Hybrid Transformer Based Feature Fusion for Self-Supervised Monocular Depth Estimation

Advances in Image Manipulation, European Conference on Computer Vision Workshops (ECCVW) 2022

• Authors: Snehal Singh Tomar, Maitreya Suin, and A.N. Rajagopalan
• Paper: ECCVW 2022 arXiv preprint

Setup

Our setup for this project entailed the following:

• CUDA 10.0, cuDNN 7.5.0, Python 3.6, Pytorch 0.4.1, Torchvision 0.2.1, OpenCV 3.3.1, and Ubuntu 20.04.
• Upon cloning the repository, please place the KITTI Dataset in "kitti_data/" before running any experiments.

Training

Please run:

python train_w_lrl_hrl.py --model_name <desired_model_name> --png

The "--png" option may be omitted if the KITTI Dataset has been downloaded in ".jpg" file format. The model files will be saved at "model_under_trg/desired_model_name/" by default.

Inference

To generate predicted depth maps, run:

python export_gt_depth.py --data_path kitti_data --split eigen

To evaluate a particular model, run:

python evaluate_depth.py --load_weights_folder <path_to_model_weights> --eval_mono

Bibtex

If you use this code, please cite our paper:

@inproceedings{tomar2022hybrid,
  title={Hybrid Transformer Based Feature Fusion for Self-Supervised Monocular Depth Estimation},
  author={Tomar, Snehal Singh and Suin, Maitreya and Rajagopalan, A.N.},
  booktitle={Advances in Image Manipulation, European Conference on Computer Vision Workshops (ECCVW) 2022},
  year={2022}
}

License

This code is for non-commercial use only. Please refer to our License file for more.

Acknowledgement

This implementation borrows heavily from Monodepth2, and draws inspiration from the DIFFNet.