Repository navigation
Releases
Virtual Path Vision v3.2.0 — Surface Defect & Agentic Inspection
Virtual Path Vision v3.2.0
Surface Defect mode and the OpenCV AI Competition 2026 pipeline are now in a release build.
This release is the first packaged build that includes the competition work. The previous v3.1.1 release predates the Surface Defect mode.
What's new
- Surface Defect mode — the 12th processing mode in the operator console. OpenCV 5 two-scale detection for scratches, spots, and low-contrast stains; NG evidence PNGs; stabilised production-line verdicts.
- Agentic inspection pipeline — synthetic 800-part benchmark, vision-driven re-inspection, JSONL tool-call traces, and reproducible evidence artifacts.
- VirtualPath-Core virtual camera — consume the engine's headless MJPEG stream on
/cam1like an industrial camera. - AWS evidence path — S3, Lambda, CloudWatch, SNS, and a CloudFormation template. The included receipt is dry-run only.
- Security and quality —
Microsoft.Data.Sqlite 8.0.33/SQLitePCLRaw 2.1.13(fixes GHSA-2m69-gcr7-jv3q), English evaluation/artifact output, and CI regression tests.
Download
VirtualPathVision-win-x64-v3.2.0.zip— self-contained Windows 10/11 x64 build; no separate .NET installation required.- SHA256:
53443b8be907d7f8f31711bb0b517c48f5862ef3dd009223bcf00a62b91c23d0
Extract the archive and run VirtualPathVision.exe. The archive also contains README.txt and the GPL-3.0 LICENSE.
Verification
- Release build: 0 warnings, 0 errors.
- Regression tests: 87 production-line checks plus the inspection/agent suite pass.
- SQLite runtime connection verified.
- Published executable smoke-tested (starts and closes cleanly).
Scope and honest limits
- The 800-part benchmark is synthetic, not a real production line.
- Defects within 17 px of the part boundary are not inspected.
- AWS is dry-run verified only; no live AWS deployment is claimed.
- Physical USB/MSMF camera and MES/PLC integration are not commissioned in this release.
Links
- Repository: https://github.com/Virtual-Path/virtual-path-vision
- Evidence artifacts: https://github.com/Virtual-Path/virtual-path-vision/tree/master/artifacts
- Technical report: https://github.com/Virtual-Path/virtual-path-vision/blob/master/docs/competition/TECHNICAL_REPORT.md
- 3D engine: https://github.com/Virtual-Path/virtual-path-core
Virtual Path Vision v3.1.1
Fix release — the app no longer fails to start when the bundled face-detection model is missing.
🐞 Fixed
- Startup crash when
face_detection_yunet_2023mar.onnxis missing or unreadable.
Previously this threw duringMainWindowconstruction and the process exited immediately
(XamlParseException → OpenCvSharp.OpenCVException: Can't read ONNX file).
Now the app starts normally and simply runs without face detection. - Unhandled exceptions now show a readable error dialog instead of disappearing silently,
so problems are visible and reportable.
📦 Download
| Platform | File |
|---|---|
| Windows 10 / 11 (x64) | VirtualPathVision-win-x64-v3.1.1.zip |
Self-contained — no .NET runtime required. Unzip anywhere and run VirtualPathVision.exe
(make sure you keep all files from the archive in the same folder).
⚠️ If it still won't start
- Extract the whole archive to a normal folder (avoid moving only the
.exe).
The app needs the files next to it (*.dll,face_detection_yunet_2023mar.onnx, …). - Run
VirtualPathVision.exefrom inside that extracted folder. - If a dialog appears, it now contains the exact error — please report it.
🔖 What's in 3.1.x
See v3.1.0 for the
full feature list (AI active perception, AWS, industrial protocols, light/dark themes, UI redesign).
Full changelog: CHANGELOG.md
Virtual Path Vision v3.1.0
OpenCV 5 · AWS Cloud · AI Active Perception — a modern Windows machine-vision desktop app.
ℹ️ This release also renames the project from Machine Vision App to Virtual Path Vision
(namespaceVirtualPathVision, repositoryvirtual-path/virtual-path-vision).
📦 Download
| Platform | File |
|---|---|
| Windows 10 / 11 (x64) | VirtualPathVision-win-x64-v3.1.0.zip |
Self-contained — no .NET runtime needed. Unzip anywhere and run VirtualPathVision.exe.
✨ Highlights
Added
- AI active perception — YOLO detection, Kalman multi-object tracking, digital-twin overlay, per-class breakdown
- AWS cloud — S3 upload, IoT Core publish, Lambda invoke
- Industrial connectivity — Modbus TCP, OPC UA, serial / TCP barcode scanners, work-report export
- Light / dark themes with runtime switching and a Follow system option
- Collapsible sidebar with persisted state, plus a dual camera preview toggle
- Responsive layout — multi-column panels stack automatically on narrow windows
- Open-source scaffolding:
LICENSE(GPL-3.0),CONTRIBUTING,SECURITY,.githubtemplates, CI
Changed
- Full UI redesign — Apple-style card layout, unified spacing, refined sliders and toggles
- Preview now scales to fit the window (no more oversized preview on high-resolution cameras)
PerMonitorV2DPI awareness — crisp rendering on 2K / 4K displays- Upgraded to OpenCvSharp5 (OpenCV 5) and AWS SDK v4
Fixed
- Settings window no longer crashes on open
- Switching language no longer wipes the saved theme preference
- The saved theme is correctly restored on startup
- Removed stray 1-px highlight lines inside buttons and inputs
✅ Requirements
- Windows 10 / 11 (x64)
- A DirectShow / MSMF-compatible camera for live capture (network / image sources work without one)
📄 Full changelog
See CHANGELOG.md.
Machine Vision App v2.2.0
新增功能
- 形状识别:自动分类圆形/矩形/三角形/五边形/多边形,彩色轮廓+统计
- ORB 特征点匹配:RANSAC 单应矩阵定位,抗旋转缩放
- 图像增强:CLAHE + 锐化
- 颜色检测取色:点击画面直接选取目标颜色
- 11 种处理模式:Canny / Sobel / Laplacian / 二值化 / 轮廓 / QR条码 / 颜色 / 模板匹配 / 形状 / 特征点 / 增强
- 人脸检测、录像、截图、网络摄像头、中英双语
安装
- 下载 MachineVisionApp-win-x64-v2.2.0.zip
- 解压后运行 MachineVisionApp.exe(无需安装 .NET,自包含)
测试图片
仓库 TestImages/ 目录自带测试图:test_scene.png(全功能)、template_green.png(模板匹配)、template_qr.png(特征匹配)、face_lena.jpg(人脸检测)