Skip to content
This is the official website of our work 3D Appearance Super-Resolution with Deep Learning published on CVPR2019.
Python Shell
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.
code Modified Apr 9, 2019
experiment Quick start Apr 9, 2019
LICENSE Initial commit Apr 2, 2019 Update Jun 4, 2019

This is the official website of our work 3D Appearance Super-Resolution with Deep Learning (arxiv) published on CVPR2019.

We provided 3DASR, a 3D appearance SR dataset that captures both synthetic and real scenes with a large variety of texture characteristics. The dataset contains ground truth HR texture maps and LR texture maps of scaling factors ×2, ×3, and ×4. The 3D mesh, multi-view images, projection matrices, and normal maps are also provided. We introduced a deep learning-based SR framework in the multi-view setting. We showed that 2D deep learning-based SR techniques can successfully be adapted to the new texture domain by introducing the geometric information via normal maps.

alt text

We introduce the 3DASR, a 3D appearance SR dataset and a deep learning-based approach to super-resolve the appearance of 3D objects.


  • Python 3.6
  • PyTorch >= 1.0.0
  • numpy
  • skimage
  • imageio
  • matplotlib
  • tqdm

Quick Start (Test)

  1. git clone

  2. Download pretrained model and texture map dataset.

  3. Put pretrained model at ./experiment/.

  4. cd ./code/script

    CUDA_VISIBLE_DEVICES=xx python ../ --model FINETUNE --submodel NLR --save Test/NLR_first --scale 4 --n_resblocks 32 --n_feats 256 --res_scale 0.1 --pre_train ../../experiment/model/NLR/ --data_train texture --data_test texture --model_one one --subset . --normal_lr lr --input_res lr --chop --reset --save_results --print_model --test_only

    Use --ext sep_reset for the first run that uses a specific split of the two splits from cross-validation.

    Be sure to change log directory --dir and data directory --dir_data.

How to Run the Code

Prepare pretrained model

  1. Download our pretrained model for 3D appearance SR from google drive. The pretrained models of NLR and NHR in the paper are included.

  2. Download the pretrained EDSR model from EDSR project page.

  3. Put the pretrained model at ./experiment.

Prepare dataset

  1. Download the texture map of the proposed 3D appearance dataset.

Train and test

  1. Please refer to for the training and testing demo script. In a batch system, you can also use
  2. Remember to change the log directory --dir and data directory --dir_data. --dir is the directory where you put your log information and the trained model. --dir_data is the directory where you put the dataset.


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

Yawei Li , Vagia Tsiminaki, Radu Timofte, Marc Pollefeys, and Luc van Gool, "3D Appearance Super-Resolution with Deep Learning" In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2019.

  title={3D Appearance Super-Resolution with Deep Learning},
  author={Li, Yawei and Tsiminaki, Vagia and Timofte, Radu and Pollefeys, Marc and Van Gool, Luc},
  booktitle={In Proceedings of the IEEE International Conference on Computer Vision},
You can’t perform that action at this time.