How to run any custom .pb model file on deepstream 5.0?

I have created custom objetc_detection model which save in object_detection.pb file format, so how to run this file on deepstream 5.0 ?

I have jetson nano, jetpack 4.4, cuda 10.2, deepstream 5.0.
please suggest steps…

1 Like

you need to try convert .pd model to .onnx model

please suggest for after conversion .pb model to .onnx model whats the next steps to run .onnx model on deepstream ?

can we run any customized trained model on deepstream or its need to be train some different way ?
if we need to be train different way so please suggest steps ?

Can Deepstream Run on RTX Series like 2060 ?

Whether or not Deepstream will give good performance on RTX 2060 card ?

Hi,

1.
You can run the ONNX model with Deepstream if TensorRT supports all the model layers.
Here is the TensorRT support matrix for your reference.
And you can run the model with trtexec to confirm like this:

/usr/src/tensorrt/bin/trtexec --onnx=[your/model]

2.
And based on our document below, you can run dGPU version Deepstream on an RTX Series:
https://docs.nvidia.com/metropolis/deepstream/dev-guide/text/DS_Quickstart.html#dgpu-setup-for-ubuntu

Although we don’t have profiling data for 2060, here is one for T4:
And you should get good performance for dGPU since Deepstream optimizes for both Jetson and dGPU.
https://docs.nvidia.com/metropolis/deepstream/dev-guide/text/DS_Performance.html

Thanks.

I want to use pre-trained model as per my requirement . my question is can i change plugin files code according to my requirement ?

I want to change plugin .cpp file in deepstream , is it possible or not ?

I run the object detection pre-trained model successfully Now i want rectangle box output parameter , please help…

1 Like

try with test5

You mean Deepstream_Python_test _5 . Py

What the change need in python code for getting detected rectangle box parameters like x, y,w, h?
Please suggest

are you working with python binding

Whenever I am starting Jetson nano hi just showing NVIDIA on display not going ahed. Please suggest…

Yes

sorry i just work with c an c++

I also tried with some Pre-trained model in c++ , trying to changing plugin files code but I was failed… Are you successfully done regarding this issue? Then please suggest me, I will try

please share deepstream sdk 5.0 version link for download and install in jetson nano developer kit.

Please find it from https://developer.nvidia.com/embedded/deepstream-on-jetson-downloads-archived

1 Like

Thanks.

How we can use peoplenet.etlt model using python on jetson nano ?