Title: PHAF: Instant Personalized Hand Avatars

Abstract:

We introduce PHAF (Personalized Hand Avatars in a Flash), a novel system for creating photo-realistic, personalized hand avatars. This method achieves high-quality multi-view rendering using only two input images: dorsal and palmar views. In contrast to traditional optimization-based approaches that are computationally slow, PHAF rapidly generates personalized textures, making it suitable for real-time deployment on edge devices.

Our technique integrates semantic-guided mesh alignment with densified texture extraction to efficiently transfer high-frequency details. Additionally, a view-based inpainting network is employed to refine these textures, ensuring a smooth and continuous visual appearance. By leveraging a parametric hand model, PHAF supports accurate articulations and generalizes well to novel viewpoints, ensuring compatibility with standard graphics engines. Experimental results demonstrate that PHAF matches the visual fidelity of existing methods while reducing texture generation time by a factor of 30, thereby facilitating practical applications in augmented and virtual reality.

Source: arXiv Generated at: 2026-06-03 00:00:00 UTC