luxonis/depthai-ml-training

mobilenetssd for OAK-1

Opened this issue · 3 comments

moradMIRO commented

I would like to know how to train a MobileNetSSD model for OAK-1 Lite. I noticed that the existing notebooks for training MobileNetSSD are deprecated. Could you please provide any methods or guidance on how to proceed with training? Thank you for your help

tersekmatija commented

@moradMIRO
Hey, in general you should get better performance from a newer architecture like YoloV6n. Any particular interest you're interested in training MobileNetSSD?

moradMIRO commented

Thank you for your reply. Currently, I am conducting a benchmark of object detection models on OAK-1 Lite. I have achieved a frame rate of 35 FPS with YOLOv6n and YOLOv8n. My objective is to test the performance of the MobileNetSSD (v2) architecture on OAK-1 Lite. My goal is to achieve a frame rate of 50-60 FPS.

tersekmatija commented

@moradMIRO
Have you considered using smaller input shapes?

We are exploring whether we can provide a Pytorch based training notebooks that would train a MobilenetSSDLite model. However, note that Yolos should be more accurate.

In the meantime, you can still try using the deprecated notebook locally. Colab removed support for TF1, which is why we deprecated the notebook. Alternatively, you can also use the notebook in a Paperspace notebook. I believe they offer some free GPUs. You'll need to install TF1.15 and other required libraries.