Hippocampus-DETR: An Explicit Memory Object Detection Framework Based on Hippocampus Modeling

A new object-detection framework called Hippocampus-DETR incorporates an explicit memory module modeled on the biological hippocampus, according to a preprint posted to the arXiv repository on June 26, 2026 [1][2]. The architecture integrates a component named HipNet into the standard DETR detection pipeline, simulating the anatomical subregions of the hippocampus to improve feature encoding [2]. The framework systematically replicates the functional organization of hippocampal subregions — including the entorhinal cortex, dentate gyrus, CA3, CA1, and subiculum — to perform pattern separation, pattern completion, importance filtering, and information integration on visual features [2]. A layer-wise training strategy optimizes the different memory submodules, yielding a system with memory retrieval and completion capabilities [2]. The authors report that Hippocampus-DETR achieves higher detection accuracy than current mainstream models [1][2]. Beyond object detection, models equipped with the framework demonstrated generalization and data efficiency in few-shot image classification, multimodal feature construction, and image restoration tasks [2]. Subsequent experiments validated the functional necessity and internal interpretability of each memory submodule [2]. The project code is publicly available on GitHub [2]. The paper appears on arXiv, an open-access repository that hosts electronic preprints across physics, mathematics, computer science, and related fields and that, as of November 2024, receives roughly 24,000 submissions per month [6]. The study has not yet undergone peer review, consistent with the standard preprint process on the platform [6]. The authors frame the work as a technical pathway for integrating neurocognitive mechanisms with deep learning models, aiming to improve learning efficiency and task robustness [2].