Finding where the object is, drawing its outline, and telling one face from another.
9h · 12 lessons · 1 challenge
Before this: Convolutional Networks
Classification says what is in the image. Everything else in vision says where. This module builds up to that: localization, landmarks, sliding windows done convolutionally, and then YOLO, with the pieces that make it work -- intersection over union, non-max suppression and anchor boxes.
Then segmentation with U-Net, and the recognition problems that need a different loss entirely: Siamese networks and the triplet loss.
One detector, four scenes and a fixed budget of anchors and thresholds. Maximize the score.
Worth 400 XP.
