Self-Correction-Human-Parsing

Why we included this project

Human parsing means labeling each pixel of a person's image as hair, face, clothing, or a specific limb, and this project gives you a working model for that out of the box. It ships trained weights for three datasets (LIP, ATR, Pascal-Person-Part) with different label sets, so you can pick the one that matches your domain. ATR suits fashion work, while LIP covers more crowded real-world scenes. The code also includes an extension for multi-person and video parsing, which helps if your images contain several people or you need consistent labels across frames. Because the model is built for single-person parsing with a self-correction mechanism, it is easiest to adopt for single-subject images and then extend as your needs grow.