In our experiments, we confirmed that excessive data augmentation led to a pronounced label-erasing effect, demonstrating significant performance disparities among various clothing categories.
The label-erasing effect varied dramatically with the classes, indicating that not all classes are equally affected by data augmentation techniques, leading to biased model performance.
#data-augmentation #bias-in-machine-learning #convolutional-neural-networks #image-classification #experimental-results
Collection
[
|
...
]