Broad intuition · Study map 17 of 85
Vision: sliding detectors and image tokens
Suggested preparation: Embeddings: useful coordinates for meaning
Picture it
A convolution slides the same small detector across an image. Early layers can respond to local patterns; later layers combine them. A vision transformer instead treats patches as tokens that exchange information.
What to understand
- Convolution, receptive field, pooling, and translation bias
- Classification, detection, segmentation, CNNs, and vision transformers
Follow the references
1 · Course
Stanford · CS231n computer vision ↗Use the public previous-year videos and useful notes on convolution and vision transformers.
You understand it when…
Why does finding every object require more than assigning one label to an image?
This is a study map. Follow the reference for the full explanation; use the question to test your mental model. Browse the reference shelf.