#visual perception

PEVA is a new AI model that predicts egocentric video frames conditioned on detailed full-body motion, improving visual anticipation in embodied systems.

ByteDance introduces VGR, a multimodal large language model that dynamically integrates visual information during reasoning, significantly improving accuracy and efficiency in vision-language benchmarks.