What's this about?
An 8B agent links image/video regions, entities, retrieved sources and answer steps through a visually grounded evidence graph. Authors report gains of 20.2 and 17.6 points over tool-enabled Qwen3-VL-8B on two newly introduced benchmarks.
What exactly is evidence graph?
A map connecting a claim to the image region, source or operation supporting it, like a detective’s board with traceable links.
What's the community saying / why does it matter?
HF: 3 upvotes; no substantive discussion collected. New author-built benchmarks are informative but need external testing to establish general performance.