Defect Detection | AI Vision

Training Model

Initializing...

Extract

Build

Calibrate

Ready

Click slots above to upload reference images.

Camera not started

Start camera and capture up to 6 snapshots of good samples.

Stream URL

Supports: RTSP, HTTP MJPEG, or snapshot URL

Enter stream URL and connect

Connect to IP camera and capture up to 6 snapshots.

Add 1-6 reference images of defect-free products to train the model.

Backbone

Pre-trained feature extractor

Feature Layers

Multi-layer for best accuracy

Detection Sensitivity: 0.7

Low (fewer false alarms) High (catch all defects)

Industrial QC: 0.6-0.8 recommended

Coreset Sampling: 50%

Fast (5%) Accurate (50%)

Higher = more accurate, slower inference

Test Image

No image

Drag & drop an image here
or browse files

Click to toggle heatmap

Analyzing image...

Awaiting Analysis

Upload reference images and a test image to start defect detection.