Estimate how many effective training samples a set of augmentation transforms actually adds, from the number of transforms and repeat factor. Use it to judge whether your dataset is large enough.
过度增强可能引入噪声。
| Item | Formula / Note | Source / Standard |
|---|---|---|
| Equivalent samples | augmentation adds fewer effective samples than the multiplier suggests | Data augmentation literature |
| Information ceiling | augmentation adds no new information, only invariance priors | Data augmentation literature |
| Evaluation effect | validation must match the real distribution or scores inflate | Machine learning practice |