https://github.com/dito97/rf-piqa
A machine learning toolkit for reference-free panorama IQA modelling.
https://github.com/dito97/rf-piqa
distillation image-quality-assessment machine-learning panorama toolkit uv
Last synced: 8 months ago
JSON representation
A machine learning toolkit for reference-free panorama IQA modelling.
- Host: GitHub
- URL: https://github.com/dito97/rf-piqa
- Owner: DiTo97
- License: mit
- Created: 2025-02-12T17:05:06.000Z (8 months ago)
- Default Branch: main
- Last Pushed: 2025-02-13T03:05:18.000Z (8 months ago)
- Last Synced: 2025-02-19T22:56:40.664Z (8 months ago)
- Topics: distillation, image-quality-assessment, machine-learning, panorama, toolkit, uv
- Language: Python
- Homepage:
- Size: 1.1 MB
- Stars: 1
- Watchers: 1
- Forks: 0
- Open Issues: 4
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Rf-PIQA
A machine learning toolkit for reference-free panorama image quality assessment (Rf-PIQA) modelling.
## 🚀 installation
```shell
python -m pip git+https://github.com/DiTo97/Rf-PIQA.git
```
## 🌟 overview
The toolkit supports two types of panorama image quality assessment (IQA):
- **reference mode:** The model sees both a high-resolution panorama and its low‐resolution constituents.
- **reference-free mode:** The model only sees a high-resolution panorama.
A fully trained reference PIQA model shall be distilled into a reference‐less model via **teacher-student** training.
The toolkit supports two types of regression head:
- A simple "value estimate" head (fully connected layer).
- A PIVEN head for prediction intervals along with the value, and its corresponding loss[^1].
The PIVEN head is more expensive at training time, but enables confidence estimates on its predictions.
## 📄 documentation
The toolkit's documentation is hosted as a [GitHub wiki](https://github.com/DiTo97/Rf-PIQA/wiki).
## 🤝 contributing
contributions to **Rf-PIQA** are welcome!
feel free to submit pull requests or open issues on our repository.
## 📄 license
see the [LICENSE](LICENSE) file for more details.
[^1]: [Simhayev et al., PIVEN: A Deep Neural Network for Prediction Intervals with Specific Value Prediction, 2020](https://arxiv.org/abs/2006.05139)