Overview
A program that finds faces in an image or video feed and classifies the expression on each one, for emotion recognition.
How it works
It runs as a two-stage pipeline:
- Face detection. OpenCV’s Haar Cascade classifier locates faces in each frame and crops them out.
- Expression classification. Each face crop is passed to ResEmoteNet, a CNN for facial emotion recognition, implemented in PyTorch, which predicts the expression class.
Splitting detection from classification keeps each stage simple: the cheap detector does the searching, and the network only ever sees a tightly cropped face.
Stack
Python · PyTorch · OpenCV