Qixing Huang

Hidden-state representation alignment in diffusion models isn't just an empirical heuristic—it is mathematically equivalent to score distillation operating directly within the latent representation space.

Consistent spherical parameterizations can be learned across diverse shapes, drastically reducing geometric distortion and enhancing model robustness.

Achieving state-of-the-art mesh quality with 18x faster inference times by directly generating triangle soups while respecting crucial symmetries.