EdgeZSAD: Practical Zero-Shot Anomaly Detection on Edge Devices
Not on the map. Why?
Abstract (arXiv)
Industrial inspection needs zero-shot anomaly detection (ZSAD) that remains useful under edge deployment constraints. Recent methods often rely on ViT-L foundation backbones (~300M parameters), which exceed the memory and operator budget of typical embedded hardware. We study this regime through EdgeZSAD, a compact reference system built around a TinyViT-21M-512 backbone, an asymmetric global-local readout (EdgeGLR), and a reproducible source-side training recipe (Real-IAD-DR). We train a single checkpoint in a source-trained, target-unseen protocol and evaluate it across six industrial benchmarks. Across three independent runs, the resulting model reaches an average image AUROC of 91.6 on MVTec-AD and 88.2 on VisA, while remaining directly deployable on Jetson Orin Nano Super (TensorRT FP16) and RB5 Gen2 (QNN GPU FP16). Across the six device-rescored benchmarks, image-AUROC drift stays below 0.2 points, indicating that the exported graph preserves host-side ranking behavior in the evaluated deployment setting.
Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 28 of its 34 references that are in the map. First posted on arXiv 15 Jun 2026.
Builds on
28 of 34 references.
References · 28
- 2021Learning Transferable Visual Models From Natural Language Supervisioncited by 33,744
- 2023DINOv2: Learning Robust Visual Features without Supervisioncited by 7,974
- 2019EfficientNet: Rethinking Model Scaling for Convolutional Neural Networkscited by 6,442
- 2019Searching for MobileNetV3cited by 1,994
- 2021EfficientNetV2: Smaller Models and Faster Trainingcited by 1,135
- 2020Designing Network Design Spacescited by 960
- 2021CrossViT: Cross-Attention Multi-Scale Vision Transformer for Image Classificationcited by 806
- 2021Towards Total Recall in Industrial Anomaly Detectioncited by 776
- 2021DRAEM -- A discriminatively trained reconstruction embedding for surface anomaly detectioncited by 514
- 2022SPot-the-Difference Self-Supervised Pre-training for Anomaly Detection and Segmentationcited by 400
- 2023WinCLIP: Zero-/Few-Shot Anomaly Classification and Segmentationcited by 315
- 2021Post-Training Quantization for Vision Transformercited by 278
16 more
- 2023AnomalyCLIP: Object-agnostic Prompt Learning for Zero-shot Anomaly Detectioncited by 271
- 2022EfficientFormer: Vision Transformers at MobileNet Speedcited by 241
- 2022TinyViT: Fast Pretraining Distillation for Small Vision Transformerscited by 234
- 2021VT-ADL: A Vision Transformer Network for Image Anomaly Detection and Localizationcited by 223
- 2023FastViT: A Fast Hybrid Vision Transformer using Structural Reparameterizationcited by 162
- 2023EfficientAD: Accurate Visual Anomaly Detection at Millisecond-Level Latenciescited by 160
- 2024Real-IAD: A Real-World Multi-View Dataset for Benchmarking Versatile Industrial Anomaly Detectioncited by 139
- 2023APRIL-GAN: A Zero-/Few-Shot Anomaly Classification and Segmentation Method for CVPR 2023 VAND Workshop Challenge Tracks 1&2: 1st Place on Zero-shot AD and 4th Place on Few-shot ADcited by 130
- 2024AdaCLIP: Adapting CLIP with Hybrid Learnable Prompts for Zero-Shot Anomaly Detectioncited by 127
- 2021Mixed supervision for surface-defect detection: from weakly to fully supervised learningcited by 101
- 2024A Unified Anomaly Synthesis Strategy with Gradient Ascent for Industrial Anomaly Detection and Localizationcited by 74
- 2023CLIP-AD: A Language-Guided Staged Dual-Path Model for Zero-shot Anomaly Detectioncited by 58
- 2025Bayesian Prompt Flow Learning for Zero-Shot Anomaly Detectioncited by 31
- 2025AdaptCLIP: Adapting CLIP for Universal Visual Anomaly Detectioncited by 29
- 2026VisualAD: Language-Free Zero-Shot Anomaly Detection via Vision Transformercited by 11
- 2026One Language-Free Foundation Model Is Enough for Universal Vision Anomaly Detectioncited by 4
Built on by
Not cited yet.