Quantifying Bias in Image Generation | HackerNoon
Briefly

Bias evaluation in image generation models using metamorphic testing to measure social biases like gender, age, and race. Collects suspicious biased images with significant differences in properties from seed images and generated images.
BiasPainter calculates gender bias score, age bias score, and race bias score to quantify biases in generated images, ensuring fairness by analyzing (seed image, generated image) pairs.
Read at Hackernoon
[
]
[
|
]