Convolutions, the architectures that mattered, detection and segmentation, and vision transformers.
12h · 9 lessons · 1 challenge
Before this: Making Deep Networks Train
Convolution as an operation, then the architectural line from LeNet through ResNet to EfficientNet, then the tasks — classification, detection, segmentation — and finally vision transformers and the CLIP-style contrastive models that connect vision to the language track.
You build a detector, which is where the abstractions become concrete.
Hand-design convolution kernels to detect edges, corners and textures, then watch a trained network arrive at yours.
Worth 500 XP.
