Get started
Install CV-Scope on Windows
The Windows installer sets up everything CV-Scope needs in one go: a private copy of Python, PyTorch for your processor or NVIDIA graphics card, the detection models you pick and the sample videos. You do not need Python, Node.js or Git, and you do not need administrator rights.
Download: CV-Scope-Setup-Windows-x64.exe (about 16 MB) from the latest release. Setup downloads the rest while it installs.
Requirements
| Windows | Windows 10 version 1809 or newer, or Windows 11, 64-bit (x64). Windows on Arm is not supported. |
| Disk space | About 2 GB when CV-Scope runs on the processor, about 5 GB with NVIDIA graphics card support (plus 2.6 GB of temporary space on the system drive while installing), plus the models you choose (6 MB to 1.6 GB) and room for your videos and recordings. Setup checks the free space before it downloads anything. |
| Memory | 8 GB; 16 GB recommended for several cameras or the larger models. |
| Internet | Needed while installing. Setup downloads about 0.5 GB for the processor build or about 3 GB for an NVIDIA graphics card, plus the models you choose. Afterwards CV-Scope works offline. |
| NVIDIA graphics card (optional) | GeForce GTX 900 series or newer. GTX 900 to RTX 40 series: NVIDIA driver 528 or newer. RTX 50 series: driver 580 or newer. The CUDA Toolkit is not needed; PyTorch brings what it uses. AMD and Intel graphics are not used for detection: CV-Scope then runs on the processor, which works on any computer but detects more slowly. |
| Nothing else | No Python, Node.js, Git or Docker. No administrator rights: everything goes into your user profile. |
1. Download and check the installer
Download CV-Scope-Setup-Windows-x64.exe from the link above. Each release
also has a SHA256SUMS.txt. To check that the download is complete and
unchanged, open PowerShell in your Downloads folder and run
(Get-FileHash .\CV-Scope-Setup-Windows-x64.exe -Algorithm SHA256).HashThe result must match the value in SHA256SUMS.txt (upper or lower case
does not matter).
2. Start it (Windows SmartScreen)
The installer is not signed with a paid code-signing certificate yet, so
Windows may show “Windows protected your PC” when you open it. Click
More info, check that the app is CV-Scope-Setup-Windows-x64.exe, then
click Run anyway. If you want to be sure first, compare the checksum as
shown above.
3. Follow the setup wizard
-
Welcome: a summary of what Setup downloads.
-
Destination: where the application goes. The default,
%LOCALAPPDATA%\Programs\CV-Scope, is right for almost everyone. -
Graphics card: Setup looks for an NVIDIA graphics card and names the one it found. Choose Use the NVIDIA graphics card when it is offered (detection is several times faster), or Use the processor only. If the driver is too old, Setup says so; update it from nvidia.com/drivers and run Setup again.
-
Components: pick a preset or tick exactly what you want.
Preset What it installs Recommended The application, the YOLO11n, s and m detectors (small, balanced and accurate), the sample videos and Claude support for the Anomaly Assistant All CV-Scope models Every detector (YOLO11, YOLOv8, RT-DETR, ONNX versions, torchvision), the tracker’s appearance model and the face and plate recognition models Minimal The application only; download models later on the Models page Optional extras: the face and plate recognition models (they are used only with a recognition licence; the InsightFace models are for non-commercial research only), a PostgreSQL driver, and a local vision model for the Anomaly Assistant, which installs Ollama and downloads the chosen model (1.7 to 6 GB).
-
Data folder: where CV-Scope keeps its database, videos, recordings, exports and downloaded models. The default is
%LOCALAPPDATA%\CV-Scope. Choose a drive with room for recordings, and avoid folders that OneDrive or Dropbox synchronise (Setup warns you): synchronising the database while CV-Scope uses it can damage it. -
Shortcuts: a desktop shortcut if you want one. The Start-menu entry is always created.
-
Ready to install: a summary with the total download size. Click Install.
-
Downloading PyTorch: the largest download, with a progress bar and an Abort button. Nothing has been installed yet at this point; after an abort or a network error you can click Install again.
-
Installing: Setup copies CV-Scope, installs Python and the packages, creates the database and downloads your models. This takes a few minutes; the line under the progress bar shows what it is doing.
-
Finished: leave Start CV-Scope now ticked and click Finish.
4. Start and stop CV-Scope
Open CV-Scope from the Start menu (or the desktop). A small CV-Scope window appears, the server starts in the background (the first start can take a minute) and your web browser opens CV-Scope at http://127.0.0.1:8420. The window shows:
- whether CV-Scope is running, the address, and how many runs are in progress;
- Open CV-Scope, which opens the browser again, and Stop / Start;
- links to the data folder and the server log.
Closing the window stops CV-Scope. If runs are in progress, it asks first. Opening the shortcut again while CV-Scope runs brings the window forward and opens the browser; it never starts a second copy. If another program already uses port 8420, CV-Scope takes the next free port and shows the address it uses.
On the first visit the browser shows CV-Scope’s own setup assistant. It checks the hardware, recommends a detector (already downloaded if you chose it in Setup), offers the sample videos with one click, and opens the Scene Builder. From there, follow the guides under Help → Guides, starting with A first tour and setup.
CV-Scope only listens on this computer (127.0.0.1). Read the guide
Deploy a study before opening it to a network.
Update
Download the new installer and run it. It keeps your data, your settings and your earlier choices, and it downloads only what changed (PyTorch is reused when the version is the same). Close the CV-Scope window first; Setup asks you to if it is open.
To switch between the processor and the graphics card, run the installer again and change the choice on the Graphics card page.
Uninstall
Open Settings → Apps → Installed apps, find CV-Scope and choose Uninstall. Uninstall removes the application and its Python environment, then asks whether to delete your data folder as well. The answer No (the default) keeps your projects, recordings and models for a later installation.
Where things are
| What | Where (defaults) |
|---|---|
| Application, Python environment | %LOCALAPPDATA%\Programs\CV-Scope |
| Data: database, videos, recordings, exports, models | %LOCALAPPDATA%\CV-Scope (the folder chosen in Setup) |
| Sample videos | %LOCALAPPDATA%\Programs\CV-Scope\samples\videos |
| Server log | logs\server.log in the data folder (the previous three starts are kept) |
| Setup log | %LOCALAPPDATA%\Programs\CV-Scope\logs\install.log |
| Settings file | %LOCALAPPDATA%\Programs\CV-Scope\.env (see installation, “Configuration”; Setup keeps your lines when it updates) |
The command line is in the Python environment. In PowerShell:
& "$env:LOCALAPPDATA\Programs\CV-Scope\.venv\Scripts\cvscope.exe" hardware
& "$env:LOCALAPPDATA\Programs\CV-Scope\.venv\Scripts\cvscope.exe" models install yolo11xTroubleshooting
- “This step did not finish” during installation. Usually the internet
connection dropped or a proxy or firewall blocked a download. Click
Retry. If you cancel, run Setup again later: it continues where it
stopped. The details are in
install.log(see the table above). - Behind a company proxy. Setup downloads PyTorch with the Windows proxy
settings. The package step and model downloads use the
HTTPS_PROXYenvironment variable: set it for your user (for examplesetx HTTPS_PROXY http://proxy.example:8080), then run Setup again. - “Setup has detected that CV-Scope is currently running”. Close the CV-Scope window (not only the browser tab), then click OK.
- The graphics card is not used. Setup lists this on its last page. Open the Hardware page in CV-Scope: if CUDA is not available, update the NVIDIA driver (see the requirements), restart the computer and run Setup again with the graphics card option.
- The browser does not open, or shows an error. Click Open CV-Scope in the CV-Scope window, or open the address it shows. If the window says CV-Scope could not start, click Server log and look at the last lines.
- An antivirus program blocks or slows the installation. Setup starts
uv.exeand Python to install packages, which some antivirus programs inspect closely. Allow CV-Scope’s folder, or install while the scan is paused, then run Setup again.
Unattended installation
For IT departments and lab computers, Setup runs silently:
.\CV-Scope-Setup-Windows-x64.exe /VERYSILENT /SUPPRESSMSGBOXES /NORESTART `
/TORCH=auto /DATADIR="D:\CV-Scope data" `
/COMPONENTS="detectors\yolo11n,detectors\yolo11s,samples" /TASKS="desktopicon" `
/LOG="$env:TEMP\cvscope-setup.log"| Option | Meaning |
|---|---|
/TORCH=auto|cpu|cu126|cu130 | PyTorch build: chosen from the graphics card (default), processor only, CUDA 12.6 (GTX 900 to RTX 40 series) or CUDA 13.0 (RTX 50 series) |
/DATADIR=path | Data folder (default %LOCALAPPDATA%\CV-Scope) |
/DIR=path | Application folder |
/TYPE=recommended|full|minimal | A preset instead of /COMPONENTS |
/COMPONENTS=... | Comma-separated: detectors\yolo11n, detectors\yolo11s, detectors\yolo11m, detectors\yolo11l, detectors\yolo11x, detectors\yolov8, detectors\rtdetr, detectors\onnx, detectors\torchvision, tracking, recognition\face, recognition\insightface, recognition\plate, samples, extras\claude, extras\postgres, localai\moondream, localai\qwen3b, localai\qwen7b |
/TASKS="desktopicon" | Create the desktop shortcut (/TASKS="" for none) |
Exit codes: 0 installed; 10 the installation failed (details in the log);
11 installed, but some optional downloads (models, sample videos, the local
vision model) failed or the graphics card cannot be used yet. Uninstall
silently with "%LOCALAPPDATA%\Programs\CV-Scope\unins000.exe" /VERYSILENT,
which keeps the data folder.
How the installer is built and released is described in installer/README.md
in the repository.
This page is docs/install-windows.md from ajm2004/CV-Scope@eaf904a
, committed 30 September 2026. The repository is the source of truth; this site renders
it.