MemLens
MemLens: Benchmarking Multimodal Long-Term Memory in Large Vision-Language Models
Xiyu Ren, Zhaowei Wang, Yiming Du, Zhongwei Xie, Chi Liu, Xinlin Yang, et al. · 2026-05-14
- Modality
- Image
- Data
- Synthetic
- Task · ground truth
- QA/Ret. · Ans.+Ev.
- Scope
- Memory-specific benchmark
What it tests
- CECross-modal evidenceDirectly evaluated
- LELong-range evidenceDirectly evaluated
- SCState changesDirectly evaluated
- CFConflicting memoriesNot evaluated
- UQUnanswerable queriesDirectly evaluated
- MAMemory-grounded actionNot evaluated
- FGFine-grained visual detailDirectly evaluated
- ETEntity trackingPartially evaluated
- VCVisual state changeDirectly evaluated
Survey notes
Multimodal evidence sessions are embedded in timestamped histories; tests cover visual recall, multi-session reasoning, updates and evidence removal.