yiranran/APDrawingGAN

Implements a hierarchical GAN architecture with local discriminators and auxiliary networks for facial landmark and background awareness, enabling fine-grained control over portrait stylization. Built on PyTorch with support for pre-trained models, the framework includes preprocessing utilities for face alignment, landmark extraction, and background masking to handle real-world photo inputs. Integrates Visdom for training visualization and provides both single-image inference and batch training pipelines with optional initialization from pre-trained weights.

809 stars. No commits in the last 6 months.