I adapted your basic Norfair+YOLO tracking example to use our DeGirum SDK for model inference instead of PyTorch.
Turned out pretty nice in my opinion! The demo shows how simple the tracker is to adapt to any inference backend.
Right now I set the default to running YOLOv5 on our ORCA AI accelerator through our Cloud Platform, so a token for an account on the platform is required.
I also added a --device option to the demo script for inference using CPU, ORCA (if present), or any other device supported by our library locally or on any device where our inference server is running. Also, while the default model is set to YOLOv5, since our platform handles post-processing, all detection models on our platform have the same output format so it is possible to change to a completely different model just by overriding the --model option.
Please check it out. If you feel like this is a useful demo, I'd appreciate if it can get upstreamed.
If there are any issues or the demo would benefit from more documentation, I'll be happy to improve it!
Hi Tryolabs,
I adapted your basic Norfair+YOLO tracking example to use our DeGirum SDK for model inference instead of PyTorch.
Turned out pretty nice in my opinion! The demo shows how simple the tracker is to adapt to any inference backend.
Right now I set the default to running YOLOv5 on our ORCA AI accelerator through our Cloud Platform, so a token for an account on the platform is required.
I also added a
--deviceoption to the demo script for inference using CPU, ORCA (if present), or any other device supported by our library locally or on any device where our inference server is running. Also, while the default model is set to YOLOv5, since our platform handles post-processing, all detection models on our platform have the same output format so it is possible to change to a completely different model just by overriding the--modeloption.Please check it out. If you feel like this is a useful demo, I'd appreciate if it can get upstreamed.
If there are any issues or the demo would benefit from more documentation, I'll be happy to improve it!
Thank you!