Adversarially-Guided Portrait Matting

Sergej Chicherin, Karen Efremyan

2023-05-04Image Matting Privacy Preserving

Abstract

We present a method for generating alpha mattes using a limited data source. We pretrain a novel transformerbased model (StyleMatte) on portrait datasets. We utilize this model to provide image-mask pairs for the StyleGAN3-based network (StyleMatteGAN). This network is trained unsupervisedly and generates previously unseen imagemask training pairs that are fed back to StyleMatte. We demonstrate that the performance of the matte pulling network improves during this cycle and obtains top results on the human portraits and state-of-the-art metrics on animals dataset. Furthermore, StyleMatteGAN provides high-resolution, privacy-preserving portraits with alpha mattes, making it suitable for various image composition tasks. Our code is available at https://github.com/chroneus/stylematte

Results

Task	Dataset	Metric	Value	Model
Image Matting	AM-2K	MAD	0.0055	StyleMatte
Image Matting	AM-2K	MSE	0.0024	StyleMatte
Image Matting	AM-2K	SAD	9.602	StyleMatte
Image Matting	P3M-10k	MAD	0.004	StyleMatte
Image Matting	P3M-10k	MSE	0.0019	StyleMatte
Image Matting	P3M-10k	SAD	6.97	StyleMatte

Related Papers

A Privacy-Preserving Semantic-Segmentation Method Using Domain-Adaptation Technique2025-07-17 Federated Learning for Commercial Image Sources2025-07-17 Transformer-Based Person Identification via Wi-Fi CSI Amplitude and Phase Perturbations2025-07-17 Privacy-Preserving Fusion for Multi-Sensor Systems Under Multiple Packet Dropouts2025-07-17 Federated Learning in Open- and Closed-Loop EMG Decoding: A Privacy and Performance Perspective2025-07-16 Safeguarding Federated Learning-based Road Condition Classification2025-07-16 A Privacy-Preserving Framework for Advertising Personalization Incorporating Federated Learning and Differential Privacy2025-07-16 ZKP-FedEval: Verifiable and Privacy-Preserving Federated Evaluation using Zero-Knowledge Proofs2025-07-15