Onnx post-processing with trafficcamnet

I’ve used tao toolkit to fine tune the trafficcamnet model using the detectnetv2.ipynb notebook. The inference API is great, but I would like to create my own inference function in python. I exported my model to onnx. The input for the model is: [float32[unk__150,3,1088,1920]]. The outputs are:
[name: output_cov/Sigmoid:0
type: float32[unk__151,1,68,120],
name: output_bbox/BiasAdd:0
type: float32[unk__152,4,68,120]]

The input dimensions for the model are 1088, 1920. So the output dimensions make sense as they are the input dimensions divided by the stride which is 16. I am now trying to convert these outputs to scores and bounding boxes. I have not found any resources for doing this in python. I understand how to use NMS to consolidate bounding boxes, but I’m not even sure hows to produce the bounding boxes from the output. Has anyone had a similar workflow with any detectnet_v2 model? Are there any resources out there for this use case? Thanks!

There is no update from you for a period, assuming this is not an issue anymore. Hence we are closing this topic. If need further support, please open a new one. Thanks

You can refer to the source code https://github.com/NVIDIA/tao_tensorflow1_backend/tree/main/nvidia_tao_tf1/cv/detectnet_v2/postprocessor

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.