Unable to interpret output of TensorRT+DetectNet


I’ve trained a DetectNet on DIGITS by following some tutorials. Exported the network (Caffe) and wanted to use TensorRT to do the inference with Python.
I had to remove the last layer from my [i]deploy.prototxt[i]. It was a Python layer for clustering. Now I don’t know how to interpret the data that comes out.

I don’t know if I made a mistake when generating the engine, or I just need to do the clustering of my network output, or if I set the input correctly.

I’m sharing my code below so you guys ca take a look:



By running this on DIGITS, I got only one bounding box, [952 320 1075 358]. But none of the outputs are even close to this.

How do I fix my current output so I can get the actual bounding boxes? Am I passing the input correctly? Since I have multiple classes, how do I get each bounding box class?

Thank you in advance.


Hi, any luck on this matter or advice about : how to parse the output of detectnet (in my case detectnet_v2) after tensorrt ?

I used the tlt toolkit, generqated .elt model => .trt engine.
When it comes to use this .trt engine, i am having problems to make sense of the outputs

What are the output of the last layers ?
(from my understanding they are 2 : of shape : 122478 and 24783.
But where can we find steps to process these outputs?

thank you in advance!