Custom Object Detection Model for Jetson nano using Deepstream

Accelerated Computing Intelligent Video Analytics DeepStream SDK
, , , ,
1

Description

I want to train my model for custom object detection. I tried doing this by going through the following video link from nvidia [Jetson AI Fundamentals - S3E5 - Training Object Detection Models - YouTube](https://Jetson AI Fundamentals - S3E5 - Training Object Detection Models)
But in the end this works with detectnet but i want to use this model in deepstream as my object detection model.
I am a newbie and didnt find much documents relating to how to do this with deepstream.
So please guide me with this…

Thank You

Environment

TensorRT Version: 7.1.3
GPU Type: NVIDIA maxwell architecture with 128 nvidia cuda cores
Nvidia Driver Version: 4.5.1
CUDA Version: 10.2
Deepstream Version: 5.1
Python Version (if applicable): 3.6

Relevant Files

Please attach or include links to any models, data, files, or scripts necessary to reproduce your issue. (Github repo, Google Drive, Dropbox, etc.)

Steps To Reproduce

Please include:

  • Exact steps/commands to build your repro
  • Exact steps/commands to run your repro
  • Full traceback of errors encountered
2

Hi,
This looks like a Jetson issue. Please refer to the below samlples in case useful.

For any further assistance, we recommend you to raise it to the respective platform from the below link

Thanks!

4

i have tried above but my requirement is to make a caffe model which will support deepstream-meta-test image sample code. i am not getting much resources on how to make and do custom training for caffe model in jetson nano. So i want some guidance in this.

5

Moving to Jetson Nano forum to get better help.

7

Hi,

Which model architecture do you use?

If you are using an SSD model, please check the following sample.
In general, you can deploy a detector with Deepstream via:

  1. Modifying the configure
  2. implementing the corresponding parser.

https://elinux.org/Jetson/L4T/TRT_Customized_Example#Custom_Parser_for_SSD-MobileNet_Trained_by_Jetson-inference

Thanks.

8

@AastaLLL

  1. I want to create a custom caffe model that can run with my deepstream app because what i found is that deepstream-test3 is only capable of running object detection for caffe model.
    i.e the sample code inside that named as image-meta-test is running caffe model.
    I am a newbie and dont have much idea about model building and didnt find much resources for making my custom model using caffe.
    So please guide me with the steps to create my cutom object detection model using my own set of images.
    And in case if you have any idea that will run this model with other architecture apart from caffe then it would be welcomed.

  2. Is it mandatory to do model training on my jetson nano or can i train my model on Linux system also.?

Thanks

9

Hi,

Deepstream can support caffe, uff (tf1.x), tlt, and ONNX format.
Please check the sample shared above. It runs the object detection model via ONNX format.

Usually, training is applied on a desktop GPU rather than Jetson.

Thanks.

11

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