Custom trained model detectNet jetson_inference

Autonomous Machines Jetson & Embedded Systems Jetson Nano
#1

Hello, I have trained a custom object detection model using this guide for my Jetson Nano 2GB jetson-inference/pytorch-ssd.md at master · dusty-nv/jetson-inference · GitHub

The problem is that I cannot load the custom trained model in my python script. Here is something that I tried:

  import jetson.inference
  import jetson.utils
  import numpy as np
  import cv2
  import time

 timeStamp = time.time()
 fpsFilt=0

 net = jetson.inference.detectNet('ssd',['--model=models/fruit/ssd-mobilenet.onnx -- 
                                      labels=models/fruit/labels.txt \
                              --input-blob=input_0 --output-cvg=scores --output-bbox=boxes'])

I get the following error in terminal:

      [TRT]    model format 'custom' not supported by jetson-inference
      [TRT]    detectNet -- failed to initialize.
      jetson.inference -- detectNet failed to load network

The error leads me to believe that we cannot load custom trained models this way, is there a way to fix this problem, so that I can load custom trained model to my script. Or any workaround that people in the community have been using? Thanks for reading, any kind of help is welcome.

#3

Hi,

The format type indicates the file format rather than the model architecture.
Based on your use case, the format is ‘onnx’ and do within the support range.

To give a further suggestion, could you help to check the model_fmt value in your environment?

Thanks.

#4

Hi @nitinkumar96, try it like this instead:

 net = jetson.inference.detectNet(argv=['--model=models/fruit/ssd-mobilenet.onnx', '--labels=models/fruit/labels.txt', '--input-blob=input_0', '--output-cvg=scores', '--output-bbox=boxes'])
#5

I have a directory with my Python code and a shell script. It also contains the directory Cones-and-Cells which contains my labels.txt and ssd*.onnx files.

The shell script works fine. It contains:

detectnet /dev/video3 --model=Cones-and-Cells/ssd-mobilenet.onnx --labels=Cones-and-Cells/labels.txt --input-blob=input_0 --output-cvg=scores --output-bbox=boxes

My Python code includes (per your suggestion):

net = jetson.inference.detectNet(argv=['--model=Cones-and-Cells/ssd-mobilenet.onnx --input-blob=input_0 --output-cvg=scores --output-bbox=boxes --labels=Cones-and-Cells/labels.txt'])

However, that line fails with:

[TRT]    detected model format - custom  (extension '.txt')
[TRT]    model format 'custom' not supported by jetson-inference
[TRT]    detectNet -- failed to initialize.
jetson.inference -- detectNet failed to load network

This makes no sense to me. Why is it trying to infer the model format from the label file? What am I doing wrong?

Thanks in advance for any enlightenment!

Michael

#6

Oddly, when I change the order of parameters to this:

net = jetson.inference.detectNet(argv=['--model=Cones-and-Cells/ssd-mobilenet.onnx --labels=Cones-and-Cells/labels.txt --input-blob=input_0 --output-cvg=scores --output-bbox=boxes'])

The message changes to this:

[TRT] detected model format - custom (extension '.txt --input-blob=input_0 --output-cvg=scores --output-bbox=boxes')

It’s like the parser didn’t recognize any of the spaces in the string.

#7

Found it! @dusty_nv: your earlier reply was a little off. You were missing the quotes to make argv an array of strings.

net = jetson.inference.detectNet(argv=['--model=models/fruit/ssd-mobilenet.onnx', '--labels=models/fruit/labels.txt', ' --input-blob=input_0', '--output-cvg=scores', '--output-bbox=boxes'])

1 Like
#8

Aha, sorry - right. Glad you got it working. Correct that in the post above.