Hongjue Li

SynCLIP resolves synonym-induced grounding inconsistencies, achieving unprecedented robustness in open-vocabulary dense perception tasks.

GILC enables state-of-the-art performance in discrete diffusion models without the computational burden of retraining, revolutionizing controllable generation tasks.

Ditch the gradients: a new particle-based method solves diffusion inverse problems using only forward passes, matching gradient-based performance.