Hello,
I'm using the Metis M.2 accelerator with Aetina board, and I'm using Voyager SDK 1.4.0. I've been able to use the Metis M.2 card by issuing a reboot after resets, and that's good enough to work with for now.

My code and model has been coming along, but I've got to the stage where I now wish to combine the models, to have a 2-stage pipeline.

The aim is to have:

(a) vehicle detection, followed by

(b) license plate detection. 

For (a), I'm using an existing detector (supplied with the SDK), which is yolov5m-v7-coco-tracker and I can successfully extract bounding box co-ords and images just as a test.

For (b), I'm using a model I had to compile from ONNX format (using compile -i)

The above is fine so far, but now I'm trying to chain them into a single vehicles-then-plates.yaml file, and that's unsuccessful. When I type ./deploy vehicles-then-plates, it silently exits, with no error. It may be a path issue, or it may be something else; I think I need help with the contents of that file.

I'm using these folders:

/axelera/voyager-sdk  - contains the SDK
/axelera/compiled  - the folder in which I typed compile -i /axelera/lp_yolov8n.onnx -o my_lp_model_compiled
/axelera/compiled/my_lp_model_compiled  - the folder that got created during the compile process
/axelera/compiled/my_lp_model_compiled/compiled_model  - contains model.json, pool_l2_const.bin, etc
/axelera/myapp   - contains the vehicle-then-plates.yaml file, and also where I will placed my Python app

Please could someone kindly take a look at the yaml file, any modifications suggested, and how can I deploy it, so that I can run create_inference_stream from the Python code? It would be appreciated loads.

Here are Microsoft OneDrive links:

my_lp_model_compiled.tar.gz   (The folder that was created using compile -i)

vehicles-then-plates.yaml (The YAML file for the 2-stage pipeline)

myapp.zip (the Python code)

lp_yolov8n.onnx  (The ONNX file if needed)

Many thanks!!