Image classification on OAK Camera taking too much time

aleksajanjic99

I have trained a yolo v5 image classification model, and I used online blob converter tool http://blobconverter.luxonis.com for converting the model to blob format in order to run it on OAK camera's VPU. I have set up a NeuralNetwork node, and an input stream to which I send an image, and take the NNData from the output stream. Fetching data from the output stream takes somewhere between 400 and 600 ms, which seems to be too long. Does anyone have an idea why it is taking so long, and what could be the possible mistakes I made in the process?

jakaskerl

Hi aleksajanjic99
It would help if you could specify the model you are using (nano, small,...). What is the input size? Could you add a MRE as well?

Thanks,
Jaka

aleksajanjic99

I used yolov5n-cls.pt as base model during classification training. Image size is 128x128. I will try and reproduce a MRE, and send it.