AI & MLRecommendedAIML-CV-010
Vision Transformers
Images as sequences of patches processed by a transformer, trading convolution's built-in bias for data and scale.
- Difficulty
- 3 / 5
- Importance
- 4 / 5
- Study time
- 1–2 h
- Kind of work
- Mixed
Why it matters
The learner can patchify an image into tokens and explain what inductive bias a ViT gives up relative to a CNN.