Search papers, labs, and topics across Lattice.
The paper introduces Background Embedding Memory (BEM), a training-free module for suppressing false positives in object detection within fixed-camera environments. BEM leverages the stable background as a prior by estimating background embeddings, maintaining a prototype memory, and re-scoring detection logits based on inverse similarity. Experiments across YOLO and RT-DETR models demonstrate that BEM reduces false positives while preserving real-time performance on surveillance datasets.
Slapping a simple, training-free module onto your object detector can dramatically cut down on false positives in fixed-camera scenarios, without sacrificing speed.
Pretrained detectors perform well on benchmarks but often suffer performance degradation in real-world deployments due to distribution gaps between training data and target environments. COCO-like benchmarks emphasize category diversity rather than instance density, causing detectors trained under per-class sparsity to struggle in dense, single- or few-class scenes such as surveillance and traffic monitoring. In fixed-camera environments, the quasi-static background provides a stable, label-free prior that can be exploited at inference to suppress spurious detections. To address the issue, we propose Background Embedding Memory (BEM), a lightweight, training-free, weight-frozen module that can be attached to pretrained detectors during inference. BEM estimates clean background embeddings, maintains a prototype memory, and re-scores detection logits with an inverse-similarity, rank-weighted penalty, effectively reducing false positives while maintaining recall. Empirically, background-frame cosine similarity correlates negatively with object count and positively with Precision-Confidence AUC (P-AUC), motivating its use as a training-free control signal. Across YOLO and RT-DETR families on LLVIP and simulated surveillance streams, BEM consistently reduces false positives while preserving real-time performance. Our code is available at https://github.com/Leo-Park1214/Background-Embedding-Memory.git
