GeoDSR

Learning Continuous Depth Representation via Geometric Spatial Aggregator

Accepted to AAAI 2023 [Paper]

Xiaohang Wang*, Xuanhong Chen*, Bingbing Ni**, Zhengyan Tong, Hang Wang

* Equal contribution

** Corresponding author

The official repository with Pytorch

• This work is for arbitrary-scale RGB-guided depth map super-resolution (DSR).
• Depth map super-resolution (DSR) has been a fundamental task for 3D computer vision. While arbitrary scale DSR is a more realistic setting in this scenario, previous approaches predominantly suffer from the issue of inefficient real-numbered scale upsampling.

Results:

Dependencies

• python3.7+
• pytorch1.9+
• torchvision
• Nvidia Apex (python-only build is ok.)

Datasets

We follow Tang et al. and use the same datasets. Please refer to here to download the preprocessed datasets and extract them into data/ folder.

Pretrained Models

• Baidu Netdisk (百度网盘)：https://pan.baidu.com/s/1e2rLQFqVHIy2ZZG922XNTA

• Extraction Code (提取码)：xu7e

• Google Drive: https://drive.google.com/drive/folders/1cIvA_AYh0fve_pDhN6timhCeN6A7MhD2?usp=share_link

Please put the model under workspace/checkpoints folder.

Train

python main.py

Test

bash test.sh

Licesnse

For academic and non-commercial use only. The whole project is under the MIT license. See LICENSE for additional details.

Citation

If you find this project useful in your research, please consider citing:

@misc{GeoDSR,
  author = {Wang, Xiaohang and Chen, Xuanhong and Ni, Bingbing and Tong, Zhengyan and Wang, Hang},
  title = {Learning Continuous Depth Representation via Geometric Spatial Aggregator},
  publisher = {arXiv},
  year = {2022}
}

Ackownledgements

This code is built based on JIIF. We thank the authors for sharing the codes.

Related Projects

Learn about our other projects

[EQSR]: high-quality arbitrary-scale image super-resolution;

[VGGFace2-HQ]: high resolution face dataset VGGFace2-HQ;

[RainNet];

[Sketch Generation];

[SimSwap]: most popular face swapping project;

[ASMA-GAN]: high-quality style transfer project;