Research
MM-IssueLoc: A Controlled Benchmark for Visual Evidence in Multimodal Repository-Level Issue Localization
Real GitHub issues routinely carry screenshots, error dialogs, rendered UI states, and logs, yet repository-level issue localization is still evaluated as text-only. MM-IssueLoc is a controlled benchmark that isolates localization from patch synthesis to measure whether visual input actually helps, hurts, or is ignored by agents. Relevant for builders of SWE agents who want to know if feeding screenshots to a bug-localization pipeline is worth the tokens.
Source
↳ Follow the thread