Computer Vision Interview Questions, issue 13, Jan 14, 2026

The Generalization Gap Trap

Senior Computer Vision Engineer interview at Google DeepMind, and the interviewer asks:

โ€œWe use heavy data augmentation (Color Jitter, 30ยฐ Rotations) during training to improve robustness. Why do we strictly disable these during validation? Doesnโ€™t that break the rule that ๐˜›๐˜ณ๐˜ข๐˜ช๐˜ฏ ๐˜ข๐˜ฏ๐˜ฅ ๐˜›๐˜ฆ๐˜ด๐˜ต ๐˜ฅ๐˜ช๐˜ด๐˜ต๐˜ณ๐˜ช๐˜ฃ๐˜ถ๐˜ต๐˜ช๐˜ฐ๐˜ฏ๐˜ด ๐˜ด๐˜ฉ๐˜ฐ๐˜ถ๐˜ญ๐˜ฅ ๐˜ฎ๐˜ข๐˜ต๐˜ค๐˜ฉ?โ€

Why disabling data augmentation during evaluation is the only way to measure real generalization.

The full answer, with the mechanism and the arithmetic, is for paid subscribers on Substack.

Read it on Substack

Get the next one

Free on Substack. Unsubscribe in one click.