A Data-centric Approach to Class-specific Bias in Image Data Augmentation: Abstract and Intro | HackerNoon
Briefly

from Hackernoon6 months ago
Our study extends the inquiry of data augmentation's class-specific bias, revealing that while models like ResNet50 show uniform bias effects, Vision Transformers exhibit varied robustness.
Hackernoonhttps://hackernoon.com/a-data-centric-approach-to-class-specific-bias-in-image-data-augmentation-abstract-and-intro
The refinement of our 'data augmentation robustness scouting' method highlights a significant reduction in computational resources—training just 112 models instead of 1860, while still capturing essential trends.
Hackernoonhttps://hackernoon.com/a-data-centric-approach-to-class-specific-bias-in-image-data-augmentation-abstract-and-intro
Read at Hackernoon
#data-augmentation #machine-learning #computer-vision #model-generalization #bias-mitigation
[
|
]