Haidong Zhu

Title: Haidong Zhu and Advances in Novel View Synthesis.

Introduction

Haidong Zhu is an inventor based in Alhambra, California. His patent work addresses novel view synthesis, a technology for generating images of a scene from viewpoints that were not included in the reference images. Zhu is associated with Microsoft Technology Licensing, LLC.

Latest Patents

Zhu is named on one patent, titled “View Synthesis Utilizing Scene-Level Features and Pixel-Level Features.” The invention combines semantic representations of an overall scene with representations of individual pixels. Transformer encoders use these features to identify relationships among objects that pixel-level representations alone may not capture.

The approach aims to improve the generalizability of Neural Radiance Fields (NeRF) to unfamiliar scenes, reducing the need to retrain the system for each scene. It also supports rendering novel views from a limited number of reference images. Zhu’s listed patent portfolio includes 1 patent.

Career Highlights

The available information highlights Zhu’s work on view synthesis and NeRF-based techniques. His patent focuses on combining scene-level and pixel-level information to support more adaptable image generation.

Collaborations

Zhu’s listed coworkers include Tianyu Ding and Tianyi Chen. The patent is associated with Microsoft Technology Licensing, LLC.

Conclusion

Haidong Zhu’s patent explores how combining scene-wide and pixel-specific semantic features can improve novel view synthesis, helping NeRF-based systems render unfamiliar scenes with fewer reference images.