基于 PatchCore 的工业异常检测项目,研究 layer2/layer3 分数融合、敏感性引导收缩与风险受控门控,并提供 Streamlit 工程化 Demo。
-
Updated
Oct 24, 2026 - Python
基于 PatchCore 的工业异常检测项目,研究 layer2/layer3 分数融合、敏感性引导收缩与风险受控门控,并提供 Streamlit 工程化 Demo。
PatchCore method for Industrial Anomaly Detection + CLIP
Production-ready visual anomaly detection from normal images, featuring PaDiM, ultra-light PatchCore, automated edge deployment, and optimized ONNX, OpenVINO, TensorRT INT8, Hailo, and KV260 support.
An unofficial implementation using Pytorch for "Anomaly Clustering: Grouping Images into Coherent Clusters of Anomaly Types". Improve the algorithm with DINO pretrained ViT. Implement algorithms based on PatchCore.
Training-free deep learning for unsupervised industrial defect detection & localization on MVTec AD, PatchCore & PaDiM, with image and pixel level AUROC and a one-click free-Colab demo.
Anomaly detection for visual defect inspection: trained on normal images only, with localisation. PyTorch + MVTec AD.
Real-time manufacturing defect detection using unsupervised PatchCore — FastAPI + Streamlit + OpenVINO
Comprehensive benchmark for anomaly detection models on MVTec AD dataset using Anomalib. Includes PatchCore, EfficientAD, FastFlow, STFPM, PaDiM with Gradio demo.
兼容 Anomalib 生态的通用异常检测推理库:一张图进去,缺陷 mask + 分数 + OK/NG 出来,一个 DLL + 一个头文件,零 OpenCV 依赖。
One-class industrial defect detection and localization (PatchCore + tiled DINOv3) under a pre-registered protocol. MVTec AD 2 server: AU-PRO 62.97, SegF1 53.76.
Industrial/medical anomaly detectors struggle with distribution drift, making fixed thresholds unreliable and causing false-positive rates to shift. Synthetic anomaly training improves data efficiency, but anomaly difficulty is rarely adapted to the model’s calibration state, limiting robustness under changing conditions.
Multi-view anomaly scoring models from scarce, unbalanced, multi-class data
Detecção de defeitos de solda em placas eletrônicas com PatchCore, treinado só com peças boas. Um modelo por região do painel, todos compartilhando um único DINOv2 na GPU: 90% menos VRAM. TCC do VC Master, PUC-Rio.
정상 제품 사진만으로 만드는 외관 검사기(이상 탐지)를 미리 정한 규칙으로 잰 프로젝트 (VisA·M2AD, PatchCore·Dinomaly, ONNX Runtime CPU 검사 서비스)
PatchCore + YOLO screw anomaly detection demo with Gradio UI (MVTec AD)
Modular 2D industrial inspection platform with a swappable method interface — two anomaly-detection methods (autoencoder + PatchCore) built from scratch, served via FastAPI + React, Dockerized. Benchmarked on MVTec AD (mean image AUROC 0.72 → 0.96).
Zero-label anomaly detection: CLIP patch embeddings + VLM normality priors guiding PatchCore, benchmarked on MVTec AD, edge-ready via OpenVINO
Industrial visual anomaly detection using PatchCore with FastAPI, Streamlit and Docker deployment.
Unsupervised maize disease detection with cross-domain evaluation on real-field PlantDoc imagery and a controlled background-sensitivity experiment. PatchCore generalizes from lab to field (AUROC 0.9995); reconstruction-based methods are 4× inflated by domain shift.
To associate your repository with the patchcore topic, visit your repo's landing page and select "manage topics."