Xiatian Zhu

Achieving precise control over human motion and camera trajectories in 3D scene animation could redefine the realism and interactivity of video generation.

AVLLMs can efficiently discard certain audio-visual tokens without sacrificing prediction quality, revealing a surprising flexibility in multimodal processing.

Forget training, just nudge your text embeddings: RGSE closes the open-vocabulary object detection gap under distribution shift by directly and efficiently adapting text embeddings at test time.

Achieve 21x faster training and 11x faster rendering in novel view synthesis from degraded images by jointly tackling enhancement and reconstruction in 3D.

By reinterpreting unsupervised anomaly detection through the lens of Retrieval-Augmented Generation, RAID achieves state-of-the-art performance by using retrieved normal samples to guide noise suppression in anomaly map generation.