minivision-ai/photo2cartoon

Implements unpaired image-to-image translation using a U-GAT-IT architecture enhanced with Soft-AdaLIN normalization and Face ID Loss to preserve identity while converting photos to cartoon style. The pipeline includes face detection, keypoint-based alignment, portrait segmentation for background removal, and ONNX model export for inference. Integrates pretrained face recognition models (MobileFaceNet) and segmentation networks, with support for both PyTorch and ONNX Runtime inference.

4,018 stars. No commits in the last 6 months.