Getting Spatial Bounding Box

bhavya_p

I currently using ROS2 Humble with depthai and am using OAK-D LR. I have my own trained model in a .blob form, but am having trouble understanding how to run my model to get bounding boxes coordinates of detected objects. The example filter of spatial_bb.launch is kind of what I want as a result. I was wondering if there was a way to just use my model instead of the default?

erik

Hi @bhavya_p ,
If it's a mobilenet/yolo model then yes, you can use that, and DetectionParser will decode/parse the NN results and then the script (spatial_bb) will display the results. Thoughts?

bhavya_p

I believe the model is Yolo_Nas. Would this work? Also, if it would work would I just put the path to the blob file in the i_nn_config_path variable in the camera.yaml file. Or would I put the path to the json file?

erik

@bhavya_p no likely not. You'll need to decode NN output yourself, examples here:
luxonis/depthai-experimentstree/master/gen2-yolo