Training own detector that is compatible with ZED's Unity plugin

Software
1

I managed to get teh custom object detection inside ZED’s Unity Plugin to work. Now I can run the provided yolov7-tiny and yolov4-tiny models inside Unity.

However, I’d like to train my own models and deploy them there. For that, I’d like to know in which format I need to provide the neural network/the weights. I see that the weight files have the .weight ending, but idk where they are coming from.

Do I need to follow, e.g. this zed-examples/object detection/custom detector/cpp/tensorrt_yolov5_v6.0 at master · stereolabs/zed-examples · GitHub, to convert my PyTorch model to a TensorRT engine, or is this different?

Thanks!

2

Hi,

That’s good to know.

I recommend you to take a look at our documentation about the Custom detection feature, there is a section about training custom models (https://www.stereolabs.com/docs/object-detection/custom-od/#object-detection-steps).

You can indeed also look at our different samples to see how it has been done.

Best,
Benjamin Vallon

Stereolabs Support

3

Will do, thank you! I’ll let you know if I run into issues.

4

The OpenCV for Unity plugin has several examples, which are adaptable.
However, as openCV for Unity does not utilize the GPU it is of not much use. Will open a separate thread for that.