Metadata-Version: 2.1
Name: giskard-vision
Version: 0.0.1b2
Summary: The expansion of Giskard into testing computer vision models
Author-email: Giskard AI <hello@giskard.ai>
Requires-Python: <3.12,>=3.10
Description-Content-Type: text/markdown
License-File: LICENSE
Requires-Dist: opencv-python
Requires-Dist: numpy

# giskard-vision

Giskard's Computer Vision Expansion with:
- Landmark Detection Support

[![Full CI](https://github.com/Giskard-AI/giskard-vision/actions/workflows/build-python.yml/badge.svg)](https://github.com/Giskard-AI/giskard-vision/actions/workflows/build-python.yml)

## Install

```
pip install giskard-vision==0.0.1b1
```

To install the repo in dev mode
```shell
git clone https://github.com/Giskard-AI/giskard-vision.git
cd giskard-vision
pdm install -G :all
source .venv/bin/activate
```

## Examples

setup dev-env and check out `examples`.

## Benchmark Datasets

- [x] 300W (https://ibug.doc.ic.ac.uk/resources/300-W/)
- [x] FFHQ (https://github.com/DCGM/ffhq-features-dataset)

## Metrics

- [x] ME: Mean Euclidean distances
- [x] NME: Normalised Mean Euclidean distances
- [x] NEs: Normalised Euclidean distance
- [x] NERFMark: Normalised Euclidean distance Range Failure rate
- [x] NERFImagesMean: Means per mark of Normalised Euclidean distance Range Failure rate across images
- [x] NERFImagesStd: Standard Deviations per mark of Normalised Euclidean distance Range Failure rate across images
- [x] NERFMarksMean: Mean of Normalised Euclidean distance Range Failure across landmarks
- [x] NERFMarksStd: Standard Deviation of Normalised Euclidean distance Range Failure across landmarks
- [x] NERFImages: Average number of images for which the Mean Normalised Euclidean distance Range Failure across landmarks is above failed_mark_ratio
