Issues with deepstream object detection model and inference on Jetson NX Xavier

I am writing to seek guidance on an issue I am experiencing with my Jetson NX Xavier device and deepstream object detection models.

Firstly, I attempted to fine-tune the deepstream object detection resnet10 model for my specific use case. However, I found that I was unable to do so effectively and would appreciate any assistance or guidance that NVIDIA can provide in this regard.

Secondly, I tried using the yolov7 model for my project, but I encountered issues when attempting to run it on multiple video streams. Specifically, when trying to run the model on four videos, I found that it was very laggy and ultimately caused my Jetson NX Xavier to restart. In contrast, the deepstream object detection resnet10 model was able to handle four videos without issue.

I am wondering if there are any ways to fine-tune the resnet10 model to better suit my needs, or if there are any optimizations or configurations I can use to make the yolov7 model run more smoothly on my device.

Any help or advice that you can provide would be greatly appreciated. Thank you in advance for your time and assistance.

Please provide complete information as applicable to your setup.

• Hardware Platform (Jetson / GPU)

• DeepStream Version

• JetPack Version (valid for Jetson only)

• TensorRT Version

• NVIDIA GPU Driver Version (valid for GPU only)

• Issue Type( questions, new requirements, bugs)

• How to reproduce the issue ? (This is for bugs. Including which sample app is using, the configuration files content, the command line used and other details for reproducing)

• Requirement details( This is for new requirement. Including the module name-for which plugin or for which sample application, the function description)

Hi, here is the information :
Hardware Platform: Jetson NX Xavier

DeepStream Version: DeepStream 6.1

JetPack Version: JetPack 5.0.2

TensorRT Version: TensorRT 8.4.1

NVIDIA GPU Driver Version: Not sure how to check this.

Issue Type : Performance issue

As I mentioned, I tried to run yolov7 tiny model on my Jetson Xavier for 4 video streams but it was lagging and slow, eventually it restarted itself due to high power consumption. I tried with the Resnet10 model from Deepstream , it is fine for 4 video stream. I am trying to find a way to work on transfer learning of the Resnet10 caffemodel but it seems like do not have the way. Can you please suggest me a method ? I want to run custom object detection model which can handle 4 video stream.

  1. which sample are you referring to? 2. what is the whole media pipeline?
  2. here are some solutions to improve performance, you can lower the fps if using rtsp source, you can modify nvinfer/nvinferserver’s interval paramater to change inference frequency.
  1. I was using the deepstream-test3 app as the baseline and developed a custom c++ app.
  2. What will be affected if I change the inference frequency ? And is there a way to fine tune the resnet10 model ?
  1. some frames will not do inference, please refer to slowly for more methods.

So is there a way to finetune/ transfer learning the deepstream pretrained resnet10 caffe model ?

what do you mean about "finetune/ transfer learning the deepstream pretrained resnet10 caffe model“? do you meet any specific deepstream problems?

Do you know in the deepstream sample application, there are some pretrained model right ?

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

yes, please refer to TAO model, there are many trainable models.

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