Video anomaly detection (VAD) has been extensively researched due to its potential for intelligent video systems. However, most existing methods based on CNNs and transformers still suffer from substantial computational burdens and have room for improvement in learning spatial-temporal normality. Recently, Mamba has shown great potential for modeling long-range dependencies with linear complexity, providing an effective solution to the above dilemma. To this end, we propose a lightweight and effective Mamba-based network named STNMamba, which incorporates carefully designed Mamba modules to enhance the learning of spatial-temporal normality. Firstly, we develop a dual-encoder architecture, where the spatial encoder equipped with Multi-Scale Vision Space State Blocks (MS-VSSB) extracts multi-scale appearance features, and the temporal encoder employs Channel-Aware Vision Space State Blocks (CA-VSSB) to capture significant motion patterns. Secondly, a Spatial-Temporal Interaction Module (STIM) is introduced to integrate spatial and temporal information across multiple levels, enabling effective modeling of intrinsic spatial-temporal consistency. Within this module, the Spatial-Temporal Fusion Block (STFB) is proposed to fuse the spatial and temporal features into a unified feature space, and the memory bank is utilized to store spatial-temporal prototypes of normal patterns, restricting the model's ability to represent anomalies. Extensive experiments on three benchmark datasets demonstrate that our STNMamba achieves competitive performance with fewer parameters and lower computational costs than existing methods.

Recognizing new categories that continually emerge during production is a significant challenge in defect detection, and deep learning methods have become the mainstream solution in recent years. However, in industrial scenarios, the scarcity of new category samples in the early phases leads to catastrophic forgetting and poor generalization in these methods. To address this, we propose a novel end-to-end Incremental Few-shot DEtection TRansformer (IF-DETR), which aims to resolve the knowledge ambiguity between old and new categories in such settings. Specifically, IF-DETR follows a two-stage paradigm involving pre-training and fine-tuning, leveraging knowledge distillation (KD) to improve the fine-tuning process while distinguishing label structures to eliminate ambiguous knowledge. We design two KD losses: Feature-level Instance Aware (FIA) loss and Logit-level Hierarchy Aligned (LHA) loss. For FIA loss, we exploit the global modeling ability of transformers to generate attention masks, thereby assigning reasonable weights to valuable distillation regions and mitigating catastrophic forgetting. For LHA loss, we decouple the teacher model’s output logits based on the supervision information, applying auxiliary losses at each layer of the decoder to refine the student model’s predictions and enhance the generalization ability for new categories. To the best of our knowledge, this is the first work to introduce transformer-based detector into incremental few-shot defect detection. We conduct extensive experiments on two benchmark defect detection datasets, and IF-DETR achieves the best performance across all 9 splits and few-shot settings, demonstrating its effectiveness.