xuan2k/adds-unpaired

ADDS-DepthNet

Contents

• Introduction
• Data
• ADDS_Paddle
• ADDS_Pytorch
• Citation

Introduction

This is the official implementation of the paper:

Self-supervised Monocular Depth Estimation for All Day Images using Domain Separation

Lina Liu, Xibin Song, Mengmeng Wang, Yong Liu, Liangjun Zhang

ICCV2021

We provide two implementations here, containing PaddlePaddle and Pytorch.

Quantitative Result

Data

For data download and preparation of Oxford RobotCar dataset, please refer to Oxford RobotCar dataset data preparation

ADDS_Paddle

For PaddlePaddle implementation, please refer to PaddleVideo implementation of ADDS

ADDS_Pytorch

- Environment Setup

Assuming a fresh Anaconda distribution. We recommend to create a virtual environment with Python 3.6.6 conda create -n ADDSDepthNet python=3.6.6 . You can install the dependencies with:

conda install pytorch==1.4.0 torchvision==0.5.0 cudatoolkit=10.1 -c pytorch
pip install opencv-python
pip install tensorboardX==1.4
pip install ipython
pip install scikit-image

- Pretrained model

You can download the pretrained model here. Then put the pretrained model in "Pretrained_model".

- Prediction for a single image

You can predict depth for a single image with:

python test_simple.py --image_path path_to_the_single_image --model_name path_to_the_pretrained_model

- Training

python train.py --batch_size batch_size --data_path path_to_the_training_data

- Evaluation

bash run_evaluation.sh

Citation

If you find our work useful in your research, please cite our paper:

@inproceedings{liu2021self,
  title={Self-supervised Monocular Depth Estimation for All Day Images using Domain Separation},
  author={Liu, Lina and Song, Xibin and Wang, Mengmeng and Liu, Yong and Zhang, Liangjun},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
  pages={12737--12746},
  year={2021}
}

Acknowledgements

Our code is based on Monodepth2