Udacity’s Computer Vision capstone. A CV pipeline that detects the face (Haar Cascades), regresses the 68 facial keypoints — eyes, nose tip, mouth corners, jawline — using a small CNN, then uses those landmarks as anchor points to overlay filters (sunglasses, masks) on top of the live webcam feed in real time.
Same idea Snapchat-style filters use under the hood: once you know where the eyes and nose are in pixel coordinates, alignment becomes an affine transform, and the filter just rides on top.
