Yu Gao

Image editing models can be significantly improved by replacing monolithic reward scores with a chain-of-thought reasoning verifier that breaks down instructions into distinct principles and evaluates the edited image against each.

Seedance 2.0 leapfrogs existing models by unifying multi-modal inputs (text, image, audio, video) into a single architecture for generating high-quality, longer-duration audio-video content.

Spatial audio cues and directional priors can be jointly learned end-to-end to significantly boost keyword spotting accuracy in noisy environments, outperforming traditional cascaded approaches.

End-to-end driving gets a reliability boost: UniUncer's unified uncertainty framework slashes trajectory error by 7% and improves driving performance by 10.8% in challenging scenarios.