Lesson 2 · Computer Vision
CNNs for Vision
7 min
You'll be able to
- Explain why CNNs are the workhorse of vision
- Understand filters and feature maps
- Connect convolutions to the vision pipeline
Convolutional layers are the backbone of vision systems. A small kernel slides across the image, producing a feature map that highlights specific patterns such as edges or textures.
Stacked convolutions learn a hierarchy: low-level edges, then shapes, then whole objects. This automatic feature learning is why CNNs replaced hand-crafted filters.
Challenge
Edge detector
Explain how a filter that highlights sharp changes in intensity would reveal the edges of a building in a photo.
Knowledge Check
CNNs for vision
CNNs automatically learn features rather than relying on hand-coded filters.
The output of sliding a filter over an image is called a:
Answer all questions to submit.