PixelEyes: Decoupling Perception and Reasoning for Pinpoint Visual Evidence Seeking

Researchers from Wuhan University introduced PixelEyes, a multi-turn visual reasoning agent that decouples reasoning and perception. PixelEyes uses mask-guided search and semantic-region breadth-first search to improve visual evidence seeking. The paper also introduces the PixelEyes-6K dataset and Pinpoint-Bench benchmark.

1sec.ai