Xianda Guo

Treating distillation targets as dynamic on-policy decisions rather than static teacher outputs prevents catastrophic error propagation when adapting compact vision-language models to out-of-distribution multimodal data.

DriveCache boosts driving video generation efficiency by intelligently caching based on planned motion, leading to significant improvements in fidelity and throughput.

VGGD achieves superior rendering quality and geometric consistency by integrating pretrained visual geometry priors into the reconstruction process.