TLT 3.0 LDNet postprocessing

Accelerated Computing Intelligent Video Analytics TAO Toolkit
1

Following are the output nodes for LDNet model trained using TLT 3.0. I am planning to serve the model using Triton server.

output [
    {
        name: "output_cov/Sigmoid"
        data_type: TYPE_FP32
        dims: [1,30,40]
    },
    {
        name: "output_bbox/BiasAdd"
        data_type: TYPE_FP32
        dims: [4,30,40]
    }

What is the post processing need to do to get the final bounding box predictions?

3

LPD model or peoplenet model is based on TLT detectnet_v2 network.
For its postprocessing, refer to Postprocessing Peoplenet