Diffportrait360

Diffportrait360

Diffportrait360 is the official PyTorch implementation of the CVPR'25 paper titled "Consistent Portrait Diffusion for 360 View Synthesis". This tool enables the generation of high-quality, fully consistent 360-degree views of human heads from single-view images, essential for immersive telepresence and personalized content creation.

Key Features:

• High-quality 360-degree View Generation: Create realistic 360-degree head models from single images.
• Dual Appearance Module: Ensures global front-back consistency for generated heads.
• Custom ControlNet: Enhances back-of-head detail generation.
• Compatibility with Accessories: Accommodates various forms including human and stylized portraits, with accessories like glasses and hats.

Benefits:

• Enables more immersive augmented reality and virtual experiences.
• Facilitates real-time rendering through the generation of NeRFs (Neural Radiance Fields).
• Outperforms existing methods in terms of view consistency and quality.

Highlights:

• Built on the established framework of DiffPortrait3D.
• Adaptable to various styles and presentation forms.
• Supported by community contributions and extensive research acknowledgments.