Pytorch implementation of the paper Super-resolution 3D Human Shape from a Single Low-resolution Image accepted at ECCV 2022. Note that the code of this repo is heavily based on PIFU. We thank the authors for their great job!

Project Page

Contents

• Environment
• Dataset creation
• Train
• Test
• Citation
• Contact
• Acknowledgments

Environment

Create a conda environment from environment.yml file:

conda env create -f environment.yml

The first line of the yml file sets the new environment's name.

Dataset creation

1. Download T-Human2.0
2. Process the .obj file to make the mesh watertight with the Fast Winding Number algorithm
3. Render the training dataset following PIFU
4. For Testing set,create two folders and named "mask_final" the folder that contains the mask of the image and "image_final" the folder that contains the RGB input images.

Train

$ python train_SuRS.py --freq_save_ply 25 --residual --dataroot {path_to_input_data} --results_path {path_to_outdir} --random_flip --random_trans --random_scale --num_samples 50000 --threshold 0.05 --b_min -0.5 -0.5 -0.5 --b_max 0.5 0.5 0.5 --sigma 0.06 --resolution 512 --loadSize {input_image_size * 2} 

Test

$ python eval_SuRS.py --freq_save_ply 25 --residual --dataroot {path_to_input_data} --loadSize {input_image_size * 2} --results_path   {path_to_outdir} --num_samples 50000 --threshold 0.05 --num_threads 6 --resolution 512 --load_netG_checkpoint_path {path_to_checkpoints}/netG_epoch_12 --b_min -0.5 -0.5 -0.5 --b_max 0.5 0.5 0.5

Citation

If you find the code useful in your research, please consider citing the paper.

@inproceedings{pesavento2022super,
  title={Super-Resolution 3D Human Shape from a Single Low-Resolution Image},
  author={Pesavento, Marco and Volino, Marco and Hilton, Adrian},
  booktitle={Computer Vision--ECCV 2022: 17th European Conference, Tel Aviv, Israel, October 23--27, 2022, Proceedings, Part II},
  pages={447--464},
  year={2022},
  organization={Springer}
}

Contacts

If you meet any problems, please contact:

• Marco Pesavento: m.pesavento@surrey.ac.uk

Acknowledgments

This research was supported by UKRI EPSRC Platform Grant EP/P022529/1 and it made use of time on Tier 2 HPC facility JADE2, funded by EPSRC (EP/T022205/1).