Autonomous Quality Control · Computer Vision AI

Find Defects.
Protect Quality.

A two-stage deep learning pipeline that detects and inspects industrial hardware components in real time — combining YOLOv8 object detection with EfficientAD anomaly analysis trained on the MVTec AD dataset.

Launch Inspector → GitHub →

#1 SCREW

Status DEFECTIVE

Score 0.8734

Engine EFFICIENTAD

System Capabilities

A complete inspection pipeline

🔍

Object Detection

YOLOv8 locates screws, nuts, and bolts with bounding-box precision. Runs at 4.3 ms per frame, returning class labels, coordinates, and confidence scores.

YOLOv8n · 94.5% mAP@50

🧠

Anomaly Engine

EfficientAD generates pixel-level anomaly heatmaps from a single forward pass. Trained on the MVTec AD screw dataset — no defect labels needed at inference.

EfficientAD · 97.89% Pixel AUC

📊

Inspection Reports

Every scan produces a structured audit trail — anomaly scores, bounding boxes, crop thumbnails, heatmap overlays, and pass/fail status per component.

Print-ready · CSV export

📦

Batch Processing

Upload 2 or more images at once and the pipeline automatically runs both YOLO detection and EfficientAD anomaly scoring on every image in a single pass.

Multi-image · Unified report

How it works

Four-stage inspection flow

STEP 01

📷

Capture

Camera snapshot or file import — single image or batch of images

→

STEP 02

🎯

Detect

YOLOv8 identifies component class and bounding box per object

→

STEP 03

⚡

Analyze

EfficientAD scores anomaly and generates pixel-level heatmap

→

STEP 04

📋

Report

Full audit report with scores, heatmaps, CSV export and print

Training Data

3,500+ images.
5 curated sources.

Our YOLO model was trained on a custom corpus spanning MVTec-AD, lab-captured images, Roboflow augmentation, synthetic collages, and hard-negative mining — see how the dataset was assembled, annotated, and validated.

MVTec-AD · 320 normal Custom Lab Captures Roboflow Augmented Synthetic Collages Hard Negatives

Explore Dataset →

Output Formats

Reports & Export

Every inspection run — single image or batch — produces a structured audit report with anomaly heatmaps, bounding boxes, and per-component pass/fail verdicts. Download as a print-ready PDF or export raw scan data to CSV for further analysis.

UI Demo · Single Scan

↓ Download PNG