Wenchao-Du / GAENet

This is the code for the work accepted by ICRA2022.

Depth Completion using geometry-aware embedding

This repo is the Pytorch implementation for our paper accepted by ICRA22 on "Depth Completion using Geometry-aware Embedding", developed by Wenchao Du, Hu Chen, Hongyu Yang and Yi Zhang at Sichuan University.

Our method is trained with kitti and NYUv2 dataset, and achieves the signficant performance gains without specifical designing.

Contents

1. Dependency
2. Data
3. Pretrained models
4. Training and testing
5. Citation

Dependency

This code was tested with Python3 and Pytorch >=1.0 on Ubuntu 16.04 and above

Data

1. For outdoor environment, downloading KITTI Depth Completion Data from their website.
2. For indoor environment, downloading NYUv2 dataset.

Format your data following the data_json and configs.

Pretrained models

The original pretrained models are put into the pretrained_models folder, you could use it for validation.

Citation

If you use this code or method in your work, please cite the following:

@article{Du2022DepthCU,
	title={Depth Completion using Geometry-Aware Embedding},
	author={Wenchao Du, Hu Chen, Hongyu Yang and Yi zhang},
	booktitle={ICRA},
	year={2022}
}