Ying Fu

Achieving a 96% reduction in computation for RAW video restoration with minimal performance loss opens new avenues for efficient video processing in real-time applications.

Achieving over 90% token reduction in 3D VLMs without sacrificing performance could revolutionize the efficiency of 3D scene understanding.

NanoMorph-3D achieves unprecedented reconstruction fidelity and speed by seamlessly integrating physics-driven modeling with advanced attention mechanisms, transforming how we analyze nanomaterials.

False alarms in change detection can be drastically reduced by learning to distinguish semantic changes from non-semantic variations through innovative self-distillation techniques.

Geometry-aware distillation enables text-driven video segmentation models to achieve state-of-the-art performance by leveraging 3D geometric knowledge, even from 2D image datasets.

AMNet achieves superior low-light video enhancement by generating auxiliary representations on-the-fly, even when critical data modalities are missing.

Unsupervised object detection can now achieve category awareness, bridging the gap with supervised methods without needing any labeled data.

Achieve high-fidelity 3D reconstructions from sparse-view electron microscopy data by adapting 3D Gaussian Splatting, enabling accurate analysis of dose-sensitive materials.