I am trying to put on my OAK-D camera a custom trained model for garbage detection. After some research, I have found an already trained yolov5 model (17 classes) publicly available that can be downloaded here:
This model works for local inference. This is an example of the predictions I got using local inference.
I followed these instructions https://github.com/luxonis/depthai-experiments/tree/master/gen2-yolo/device-decoding to convert this yolov5 model into IR format and "put" it on the camera for real-time inference. I converted the .pt file using http://tools.luxonis.com/ (model shape = 640 352) and run the "python main.py" command.
However, I am not getting any predictions with the camera. Could you please help me on this?
Why is the model working on local images/videos but not on the camera for real-time inference? Is there something wrong with the model conversion to IR format?
Thanks in advance for your answer,