ZED Diagnostic | StereoLabs

ZED Diagnostic is a tool designed to help you troubleshoot issues with your ZED camera. It provides detailed information about the camera’s hardware and software status, allowing you to identify potential problems and ensure optimal performance.

How to use ZED Diagnostic

Connect your ZED camera to your computer and launch the ZED Diagnostic tool from the ZED SDK installation directory or through the Start menu (Windows) or terminal (Linux).

Under Windows, you can find ZED Diagnostic in the Start menu under the ZED SDK folder. You can also run it from the command line by navigating to the ZED SDK installation directory: C:\Program Files (x86)\ZED SDK\tools\ZED Diagnostic.exe
Under Ubuntu, you can run ZED Diagnostic from the terminal by typing ZED_Diagnostic or by navigating to the ZED SDK installation directory: /usr/local/zed/tools/ZED_Diagnostic.

After pressing the Start button, the tool will automatically detect the connected ZED camera and start gathering diagnostic information.

You can monitor the progress of the diagnostic tests in real-time. The tool will perform various checks, including:

ZED SDK: SDK and CUDA versions.
PROCESSOR: CPU processor model and motherboard information.
GRAPHICS CARD: GPU model.
CUDA OPERATIONS: CUDA operations test on each GPU.
AI MODELS: Available AI models and optimization status.
CAMERA TEST: Camera hardware tests, including model and firmware version.
DEVICES: Connectivity status.

Once the diagnostic process is complete, the tool will display a summary of the results, indicating whether the camera and system are functioning correctly.

In case of any issues, the tool will display error messages and clear diagnostic information.

After the diagnostic process completes with failures, press the Save button to export the diagnostic report as a JSON file. You can share this file with the StereoLabs support team for further assistance. Please email it to support@stereolabs.com and include a detailed description of the issue you are experiencing.

Check if the AI models are optimized

The quickest way to check which AI models are optimized on your system is to run ZED Diagnostic. No camera is required: without a camera, the CAMERA TEST item fails and the tool reports ERRORS OCCURRED, but the AI MODELS check still runs.

ZED Diagnostic results without a connected camera

When the diagnostic is complete, expand the AI MODELS item to display the status of each model:

is optimized, with a check mark: the model is downloaded and optimized, ready to be used by the ZED SDK.
is not optimized, with an information icon: the ZED SDK downloads the model, if needed, and optimizes it the first time an application uses it, which delays the application start.

ZED Diagnostic AI models optimization status

If no display is connected to the system, run the diagnostic in command line mode with ZED_Diagnostic -c. The AI Models Diagnostic section of the output lists the status of each model:

$ ZED_Diagnostic -c
ZED Diagnostic Tool
- Running ZED SDK Diagnostic : OK
    ZED SDK version: 5.5.0
    CUDA version: V13.4.92
    CUDA version: V13.2.86
- Running Processor Diagnostic : OK
    Processor:   13th Gen Intel(R) Core(TM) i7-13700KF
    Motherboard:  PRIME Z690-P D4, ASUSTeK COMPUTER INC.
- Running Graphics Card Diagnostic : OK
    Graphics card:  NVIDIA GeForce RTX 4070
- Running CUDA Operations Diagnostic : OK
    GPU 0 (NVIDIA GeForce RTX 4070): CUDA operations working correctly
- Running AI Models Diagnostic : OK
    HUMAN BODY MEDIUM DETECTION is not optimized
    HUMAN BODY ACCURATE DETECTION is not optimized
    HUMAN BODY 38 FAST DETECTION is not optimized
    HUMAN BODY 38 MEDIUM DETECTION is not optimized
    HUMAN BODY 38 ACCURATE DETECTION is not optimized
    PERSON HEAD DETECTION is not optimized
    PERSON HEAD ACCURATE DETECTION is not optimized
    REID ASSOCIATION is not optimized
    NEURAL DEPTH is not optimized
    NEURAL PLUS DEPTH is not optimized
    NEURAL DEPTH INT8 is not optimized
    MULTI CLASS DETECTION is optimized
    MULTI CLASS MEDIUM DETECTION is optimized
    MULTI CLASS ACCURATE DETECTION is optimized
    HUMAN BODY FAST DETECTION is optimized
    NEURAL LIGHT DEPTH is optimized
- Running USB Camera Diagnostic : Failed
    No Camera detected  Make sure the camera is plugged in.
- Running USB Camera Diagnostic : 50%
Diagnostic finished, a file has been generated : ZED_Diagnostic_Results.json

To optimize the models that are not optimized yet, see Optimize the AI models.

Optimize the AI models

The ZED Diagnostic tool can also help you optimize the AI models used by the ZED camera. This is particularly useful if you have recently updated the ZED SDK or if you want to ensure that the AI models are already optimized before using them with a ZED SDK application.

Click the AI MODELS button to open the Model Optimizer window. It lists the available models: the switch of each optimized model is enabled and labeled OPTIMIZED.

The AI MODELS button will only be enabled after completing a diagnostic test.

Enable the switch of each model you want to optimize, or the OPTIMIZE ALL switch to select all of them, then click OPTIMIZE SELECTION. The tool downloads and optimizes the selected models, so they are ready to be used by the ZED SDK.

The downloaded and optimized models are stored in /usr/local/zed/resources/ on Linux and in C:\ProgramData\StereoLabs\resources\ on Windows.

You can also click on CLEAR ALL MODELS to remove all downloaded and optimized AI models. This is useful if you want to free up space or reset the optimizations if something is not working properly.

Advanced Features

The ZED Diagnostic tool provides several advanced features to help you diagnose and troubleshoot issues with your ZED camera.

Under Ubuntu, open a terminal and run the following command to obtain a list of advanced commands:

$ ZED_Diagnostic --help
Usage:
    -d | Auto start the diagnostic (only for GUI mode)
    -c | Start the application in command line (Useful when no GUI is available)
    -aid | Download all models for the AI module
    -aio | Optimize (and download if needed) all models for the AI module
    -aic | Clear all models for the AI module
    -ais X | Download and Optimize the associated sl::DETECTION_MODEL (int)
    -nrlo | Download and Optimize the NEURAL depth mode
    -nrlo_plus | Download and Optimize the NEURAL PLUS depth mode
    -nrlo_light | Download and Optimize the NEURAL LIGHT depth mode
    -nrlo_all | Download and Optimize all NEURAL depth modes
    -onnx <path> [WxH] | Optimize a custom object detection ONNX model
    -r | Recover an MCU module with not valid serial number
    -dmesg | Logs kernel/driver dmesg
    -h | Display this help

These commands allow you to perform various tasks without needing to interact with the GUI. Here’s a brief explanation of each command:

-d allows you to autostart the diagnostic process without pushing any button in the GUI.
-c starts the application in command line mode, which is useful when no display is available.
-aid connects to our servers to download all the AI models, including the NEURAL depth models, without optimizing them. Use it to prepare the models for a system without an internet connection (see Offline Deployment).
-aio optimizes (and downloads if needed) all models for the AI module.
-aic clears all the downloaded and optimized AI models. This is useful if you want to free up space, or reset all the optimizations if something is not working properly.
-ais X downloads and optimizes the associated sl::DETECTION_MODEL (where X is an integer representing the model ID).
-nrlo downloads and optimizes the NEURAL depth mode.
-nrlo_plus downloads and optimizes the NEURAL PLUS depth mode.
-nrlo_light downloads and optimizes the NEURAL LIGHT depth mode.
-nrlo_all downloads and optimizes all AI-based depth modes.
-onnx <path> [WxH] optimizes a custom object detection ONNX model ahead of time, so an application calling enableObjectDetection() with OBJECT_DETECTION_MODEL::CUSTOM_YOLOLIKE_BOX_OBJECTS, CUSTOM_RFDETRLIKE_BOX_OBJECTS or CUSTOM_BOX_OBJECTS_AUTODETECT starts instantly instead of optimizing the model on first use. The optional input size (default 512x512) must match the ObjectDetectionParameters::custom_onnx_dynamic_input_shape used by the application. Available since ZED SDK 5.5.
-r recovers an MCU module with an invalid serial number (to be used with ZED Mini, ZED 2 or ZED 2i cameras).
-dmesg retrieves the kernel/driver dmesg logs, for debugging GMSL2 cameras. Read here for more information.