Segmentation using sample python application

Accelerated Computing Intelligent Video Analytics DeepStream SDK
1

Please provide complete information as applicable to your setup.

•Hardware Platform (Jetson / GPU) : T4
•DeepStream Version : 5.0
•TensorRT Version : 7
•Cuda Version: 10.2
•Cudnn Version: 7.6.5
•Ubuntu Version: 18.04
•NVIDIA GPU Driver Version: 440.33.01

Hi, I am currently working on a segmentation problem using Deepstream and wants to build the pipeline using python.

So I modified the Deepstream-test1-rtsp-out sample application pipeline, and the current pipeline looks like this:

file-source → h264-parser → nvh264-decoder → nvinfer → nvsegvisual → nvvidconv → nvosd → nvvidconv_postosd → caps → encoder → rtppay → udpsink

Basically, I have added nvsegvisual plugin in the pipeline. But I do not get a proper output (pixel size reduced significantly).

I used the Industrial and Semantic models and configuration files. I have attached a screenshot of the output that I get.

Screenshot from 2021-01-14 19-07-32

Would like to know what extra changes need to be done. Also if there are any guides to build a segmentation pipeline in python in Deepstream.

Thank You!

3

I think the output is correct per your screenshot

4

Hi @bcao, thanks for the reply. But the output video size decreases as well. Is it designed to be like that?

Also even for a single class model, multiple colour masks can be observed, just like we can see in the screenshot.

5

Yeah, the output size will be same as your model’s output for nvsegvisual and you can config the width/height by setting the plugin’s corresponding property.

there will be a background color I think, what do you expect for a single class segmentation model?