patchcore-resnet50

by amazon

Memory-bank anomaly detection achieving 99.6% AUROC on manufacturing defects

180Kdl/month

25M (ResNet-50 backbone)params

Source Use in Pipeline

Identifiers

Model ID

amazon/patchcore-resnet50

Feature URI

mixpeek://image_extractor@v1/amazon_patchcore_r50_v1

Overview

PatchCore solves cold-start anomaly detection in industrial manufacturing using only normal (non-defective) images. It builds a maximally representative memory bank of nominal patch-level features from ImageNet-pretrained models, then uses nearest-neighbor outlier detection.

On Mixpeek, PatchCore enables visual quality inspection — upload examples of normal products, and detect defects, anomalies, and deviations automatically.

Architecture

Builds a coreset memory bank of mid-level patch features from a frozen ResNet-50 (ImageNet-pretrained). Uses greedy coreset subsampling for efficient memory. Anomaly scoring via nearest-neighbor distance to the memory bank.

Mixpeek SDK Integration

import { Mixpeek } from "mixpeek";

const mx = new Mixpeek({ apiKey: "API_KEY" });

await mx.collections.ingest({
  collection_id: "my-collection",
  source: { url: "https://example.com/product-image.jpg" },
  feature_extractors: [{
    name: "anomaly_detection",
    version: "v1",
    params: { model_id: "amazon/patchcore-resnet50" }
  }]
});

Capabilities

99.6% AUROC on MVTec AD benchmark
Cold-start: only needs normal images, no defect examples
Pixel-level anomaly localization maps
Halved the error of previous best methods

Use Cases on Mixpeek

Manufacturing defect detection and quality inspection

Visual anomaly detection with only 'good' training examples

Surface inspection for scratches, dents, discoloration

Automated QA in production pipelines

Benchmarks

Dataset	Metric	Score	Source
MVTec AD	Image AUROC	99.1%	Roth et al., 2022 — Table 1
MVTec AD	Pixel AUROC	98.1%	Roth et al., 2022 — Table 1

Performance

Input Size224×224 px

GPU Latency~8ms / image (A100)

CPU Latency~90ms / image

GPU Throughput~125 images/sec (A100)

GPU Memory~0.6 GB

Requires a reference corpus of normal images for comparison

Common Pipeline Companions

openai/clip-vit-large-patch14

Visual search for normal references

Specification

FrameworkPyTorch

Organizationamazon

FeatureAnomaly Detection

Outputanomaly score + map

Modalitiesimage

RetrieverAnomaly Filter

Parameters25M (ResNet-50 backbone)

LicenseApache 2.0

Downloads/mo180K

Research Paper

Towards Total Recall in Industrial Anomaly Detection

arxiv.org

Build a pipeline with patchcore-resnet50

Add this model to a processing pipeline alongside other extractors. Combine with retrieval stages for end-to-end search.

Open Pipeline Builder