This is the official implementation of the paper Space-Angle Super-Resolution for Multi-View Images, ACM MM21
Blue and yellow frustums indicate real and virtual views respectively. Our method can up-sample both the spatial and angular resolutions for multi-view images
This code uses Ubuntu 16.04.4, CUDA 10.1 and the following Python packages
pytorch=1.3.0
torchvision=0.4.1
tensorboardx=2.4
imageio=2.9.0
scikit-image=0.18.1
Our code borrows from FreeViewSynthesis, please build the Python extension needed for preprecessing,
cd ext/preprocess
cmake -DCMAKE_BUILD_TYPE=Release .
make
We borrow Forward warping from Forward warping extension and implement our novel forward depth warping stragegy in Forward_Warping_min. Please follow their install.sh to build this extension.
We use the Deformable Convolution implementation from the DCNv2. Please follow their instruction to build this extension.
We preprocess two public Dataset Tanks and Temples, ETH for training and evaluation. For ETH, we use 5 scene:delivery_area, electro, forest, playground, terrains. You can download our preprocessed version from baidudisk, code : chvb.
You can download our pretrained depth-only and full model from depth-only, full model. Place them in 'exp/experiments'
Prepare dataset and updapte config.py with your own path
cd exp
# set train=True in config.py
python exp.py --cmd retrain --iter last --eval-dsets tat --eval-scale 0.5
cd exp
# set train=True in config.py for evaluating Tanks and Temples dataset
# set train=False in config.py for evaluating ETH dataset
python exp.py --cmd eval --iter last --eval-dsets tat --eval-scale 0.5
You can use metrics.py to calculate PSNR, SSIM and LPIPS. Results will be saved in 'exp/experiments'. Remember to update your own path.
python metrics.py --dataset Tanks_and_Temples (ETH)
If you find this code useful, please cite our paper:
@inproceedings{sun2021space,
title={Space-Angle Super-Resolution for Multi-View Images},
author={Sun, Yuqi and Cheng, Ri and Yan, Bo and Zhou, Shili},
booktitle={Proceedings of the 29th ACM International Conference on Multimedia},
pages={750--759},
year={2021}
}
Our code borrows from FreeViewSynthesis. Thanks for their excellent work.