Xihu University's Xiu Yuliang: Digital Human Reconstruction Will Gradually Become a Fine-tuning Task for Basic Models | GAIR 2025
Analysis
The article reports on the latest advancements in digital human reconstruction presented by Xiu Yuliang, an assistant professor at Xihu University, at the GAIR 2025 conference. The focus is on three projects: UP2You, ETCH, and Human3R. UP2You significantly speeds up the reconstruction process from 4 hours to 1.5 minutes by converting raw data into multi-view orthogonal images. ETCH addresses the issue of inaccurate body models by modeling the thickness between clothing and the body. Human3R achieves real-time dynamic reconstruction of both the person and the scene, running at 15FPS with 8GB of VRAM usage. The article highlights the progress in efficiency, accuracy, and real-time capabilities of digital human reconstruction, suggesting a shift towards more practical applications.
Key Takeaways
“Xiu Yuliang shared the latest three works of the Yuanxi Lab, namely UP2You, ETCH, and Human3R.”