What is Test-Time Augmentation (TTA)?
easyAnswer
- At inference, apply K augmentations (crops, flips, color jitter) to the same input, run K forward passes, and average the predictions.
- Better probability estimates, often 0.5-2% accuracy gain in classification and detection, at K× inference cost.
- Popular in Kaggle competitions and medical imaging.
- Not free — production usually skips it or uses only 2-4 augmentations.
Check yourself — multiple choice
- Augment training only
- Apply K augmentations at inference, average predictions — 0.5-2% gain at K× cost
- TTA modifies the loss
- TTA replaces training augmentation
TTA: K augmented forward passes at inference, average predictions.
#augmentation
Practise Deep Learning
214 interview questions in this topic.