DeepImageSearch
DeepImageSearch: Benchmarking Multimodal Agents for Context-Aware Image Retrieval in Visual Histories
Chenlong Deng, Mengjie Deng, Junjie Wu, Dun Zeng, Teng Wang, Qingsong Xie, et al. · 2026-05-29
- Management
- T2 Dynamic Memory
- Memory changes during deployment through a stable management policy.
- Representation
- Hybrid
- Distinct operative representation paths are explicitly linked for downstream retrieval, reasoning, or control.
How experience becomes memory
Each path is one memory component, traced from what is stored to how it is found again.
raw visual history
- Represented asRetained perceptual source evidenceSource- and modality-specific
- Stored inHierarchical multi-tier store
- Retrieved byMetadata / spatiotemporal retrieval, Planner / tool-assisted retrieval
- Source evidenceSource directly available
The agent navigates raw visual histories with fine-grained tools to recover context-dependent target images.Full paper: Memory Mechanisms. We design two complementary · PDF pp. 6, 15 · Figure 1
dual-memory navigation state
- Represented asSpatial, temporal & cross-modal structured memoryCross-represented explicit
- Stored inHierarchical multi-tier store
- Retrieved byPlanner / tool-assisted retrieval
- Source evidenceSource linked & recoverable
A dual-memory system preserves long-horizon exploration state while the agent plans multi-step retrieval.Full paper: Memory Mechanisms. We design two complementary · PDF pp. 6, 15 · Figure 1
How memory changes (T2 operations)
Update · Rule-based update
During operation, DeepImageSearch applies fixed-rule insertion or replacement to raw visual history; the memory state changes while the governing update policy remains fixed.
Consolidation · Episodic / hierarchical summarization
During operation, DeepImageSearch applies episodic or hierarchical summarization to raw visual history; the memory state changes while the governing update policy remains fixed.