Skip to main content

Not showing bounding box prediction + no GPU activity

Planned
Erwin Yonata

Hello, I want to ask how to showing my bounding box of prediction. I already activated all the required 

Environment

  • Nx Witness Version: Desktop Client version 6.0.0.38488 R4 (7cc6b3c72284).
    Built for windows-x64 with msvc.
    OpenGL version: 4.6.0 - Build 30.0.101.3111.
    OpenGL renderer: Intel(R) UHD Graphics.
    OpenGL vendor: Intel.
    OpenGL max texture size: 16384.
  • Client OS: Windows 11 Home Single Language - 23H2 Build 22631.3593
  • Server OS: Ubuntu 22.04.3 LTS
  • Server GPU: Dual tesla P4 8GB (Nvidia driver: 535.161.08) 
  • Network Topology: 1 servers, 1 cameras (rtsp streamed using mediamtx), 1 or more clients
  • Special features: None

Reproduction Scenario

  1. Download yolo model YOLOv8
  2. Export the model to onnx using sclblonnx python package. Following this instruction https://nx.docs.scailable.net/for-data-scientists/importing-models/common-models/yolov5-and-yolov8
    Call bash export-to-onnx.sh with the appropriate arguments (in particular the first and last arguments: model name, and input size).
    Update the complete_onnx.py script to add the classes your model is trained on.
  3. Upload the model to the Nx AI plugin admin server https://admin.nx.scailable.dev/system/<id>
  4. Assign to our server
  5. Open our device settings, enable the plugin and let the Nx AI plugin installing
  6. After long installation, we apply the plugin to the model that we've uploaded
  7. The result is nothing

GPU activity on the server (no activity)

 

 

0

Comments

4 comments

Date Votes
  • Ichiro
    • Network Optix team

    Hi Erwin Yonata
    Can you confirm if you have enabled the recording of the device?

    0
  • Erwin Yonata

    Already activated the recording, still doesn't work.

    Even I also try to restart the server, then enabled the recording of the device and enabled the plugins also didn't work.

    just a question, Is there any like requirement for the GPU installation like driver (must), CUDA, cuDNN, etc. or something else? 

    0
  • Josef Joubert
    Edited

    Hi thank you for your question.

    To answer your last question, you do need to have the CUDA Toolkit installed. Since you have the nvidia-smi command, I assume you have the right software. It is not necessary to install cuDNN, since the NXAI Manager will ship with the libraries it needs.

    To move on to why your model isn't working, what happens when you run the default model? Do bounding boxes show up then? Do you see GPU usage? We need to know if the problem is with your installation or with the model.

    If the default model runs but you still see no GPU usage, it's possible it didn't detect you have a CUDA runtime. Since you say the installation took a long time, I believe it did detect CUDA ( the CUDA install is a large download and it also takes a few seconds to initialise when starting the runtime ), however you can also try the manual installation: https://nx.docs.scailable.net/nx-ai-manager/7.-advanced-configuration/8.1-nx-ai-manager-manual-installation  , this should ask if you want to install the CUDA runtime or not.

    If the default model runs and you do see GPU usage, then the problem is likely your model. As always, double check requirements. If it still doesn't work, feel free to share your model so we can have a look.

    0
  • Erwin Yonata

    Thanks for the response!

    I would like to know whether the model I sent is compatible to run on NVIDIA or not?

    ONNX SCLBL 

    https://drive.google.com/file/d/1zGK0C6EDzkNc3dpZJFn-hpRRXQXew2QD/view?usp=sharing

     

    Can you help check for the compatibility of this model?

    Thank you. Appreciated

    0

Please sign in to leave a comment.

Didn't find what you were looking for?

New post