Hi, I was able to quantize/compress and then convert the my custom yolov8 trained model using the steps from Sony pointed out by @naushir in one of the threads on this forum.
But after conversion and deployment to the camera I'm noticing that my model doesn't detect any objects although it was working fine prior to conversion and the accuracy and recall were within good limits.
If you can share some details on your steps and how much training data you used etc as well as epochs/basic size etc you used and any other insights it would be if great help. Also did you use Google colab for this or some other cloud based environment versus local CPU/GPUs. I completed my training on colab but quantization etc didn't work so had to do it locally..
Thx again.
But after conversion and deployment to the camera I'm noticing that my model doesn't detect any objects although it was working fine prior to conversion and the accuracy and recall were within good limits.
If you can share some details on your steps and how much training data you used etc as well as epochs/basic size etc you used and any other insights it would be if great help. Also did you use Google colab for this or some other cloud based environment versus local CPU/GPUs. I completed my training on colab but quantization etc didn't work so had to do it locally..
Thx again.
Hi everyone,
I am using the new Raspberry Pi AI Camera connected to a Raspberry Pi 5.
I successfully deployed a custom YOLO object detection model on the camera, but now I would like to send images stored on my PC directly to the processor inside the IMX500 for benchmarking purposes. Specifically, I want to evaluate the model's accuracy and performance.
Additionally, I need to retrieve the inference times for the network deployed on the chip.
Is there a way to achieve this?
Thank you for your support!
Best regards,
G
Statistics: Posted by mhpetiwala — Sat Nov 23, 2024 2:36 pm
