giskard-vision 0.0.1b1

The expansion of Giskard into testing computer vision models

giskard-vision

Giskard's Computer Vision Expansion with:

• Landmark Detection Support

Setup

prod-env

pdm install --prod
source .venv/bin/activate

dev-env

pdm install -G :all
source .venv/bin/activate
pre-commit install

Examples

setup dev-env and check out examples.

Benchmark Datasets

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

Metrics

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