[AAAI 2025] FAOR: Fast Omni-Directional Image Super-Resolution: Adapting the Implicit Image Function with Pixel and Semantic-Wise Spherical Geometric Priors

This is the official implementation for the paper FAOR: ["Fast Omni-Directional Image Super-Resolution: Adapting the Implicit Image Function with Pixel and Semantic-Wise Spherical Geometric Priors"]

Contents

• Introduction
  • Overview
  • Visual Results
• Requirements and dependencies
• Data Preparation
• Model
• Testing
• Training

Introduction

We proposes a new ODI-SR model characterized by its capacity to perform Fast and Arbitrary-scale ODI-SR processes, denoted as FAOR. The key innovation lies in adapting the implicit image function from the planar image domain to the ERP image domain by incorporating spherical geometric priors at both the latent representation and image reconstruction stages, in a low-overhead manner. Specifically, at the latent representation stage, we adopt a pair of pixel-wise and semantic-wise sphere-to-planar distortion maps to perform affine transformations on the latent representation, thereby incorporating it with spherical properties. Moreover, during the image reconstruction stage, we introduce a geodesic-based resampling strategy, aligning the implicit image function with spherical geometrics without introducing additional parameters. As a result, the proposed FAOR outperforms the state-of-the-art ODI-SR models with a much faster inference speed.

Overview

Visual Results

Requirements and dependencies

• Python 3.8 (recommend to use Anaconda)
• Pytorch 2.1.2
• basicsr, pyyaml, tqdm, opencv-python, numpy

Installation

conda create -n faor python=3.8
conda activate faor

pip install -r requirements.txt

Data Preparation

• We use the LAU dataset, apply the same cleaning procedure as OSRT, and generate segmentation maps using Trans4PASS.
• For convenience, we have initially provided the test set of the ODI-SR dataset for testing.
• Put the datasets into the data folder as follows:

data
├── odisr
│   ├── test
│   │   ├── HR
│   │   └── map
│   ├── train
│   │   ├── HR
│   │   └── map
│   └── validation
│       ├── HR
│       └── map

Model

• Download the pre-trained model:

Model	Download
FAOR	MEGA

mkdir -p save/pretrained_models

• Put the checkpoint into the save folder as follows:

save
├── pre-trained_models
│   └── train-SEG-S4-I128-B6.pth

Testing

Given HR ODI, test FAOR set for scale x2, x4, x8 and x16. You can specify scaling factor in test config.

python test_auto.py --test_config configs/test-configs/test_ODI-SEG-SR.yaml --gpu 0

Visual results will be saved in vis_res/odisr as follows:

vis_res
├── odisr
│   ├── X2
│   ├── X4
|   ├── X8
│   └── X16

Training

sh train.sh

Contact

If you have any questions, please feel free to contact us at 2310543026@email.szu.edu.cn

Citation

If this code is useful for your research or project, we would appreciate it if you could cite our paper. The BibTeX citation is as follows:

@article{shen2025fast,
  title={Fast Omni-Directional Image Super-Resolution: Adapting the Implicit Image Function with Pixel and Semantic-Wise Spherical Geometric Priors},
  author={Shen, Xuelin and Wang, Yitong and Zheng, Silin and Xiao, Kang and Yang, Wenhan and Wang, Xu},
  journal={arXiv preprint arXiv:2502.05902},
  year={2025}
}

License

This project is released under the MIT license.