RefMem-Bench
Connecting the Dots: Benchmarking Reflective Memory in Long-Horizon Dialogue
Jingjie Lin, Bingbing Wang, Zihan Wang, Zhengda Jin, Weiming Qiao, Jing Li, et al. · 2026-05-31
- Modality
- Image
- Data
- Synthetic
- Task · ground truth
- QA/Gen. · Ans.+Ev.
- Scope
- Memory-specific benchmark
What it tests
- CECross-modal evidenceDirectly evaluated
- LELong-range evidenceDirectly evaluated
- SCState changesPartially evaluated
- CFConflicting memoriesNot evaluated
- UQUnanswerable queriesNot evaluated
- MAMemory-grounded actionNot evaluated
- FGFine-grained visual detailPartially evaluated
- ETEntity trackingPartially evaluated
- VCVisual state changePartially evaluated
Survey notes
Long-horizon dialogue and speaker-level image histories support reflective-memory questions; answer accuracy and evidence-memory recall are evaluated.