Low accuracy for analytics line crossing (Falce negative)

Accelerated Computing Intelligent Video Analytics DeepStream SDK
1

Please provide complete information as applicable to your setup.

• Hardware Platform (Jetson / GPU) NVIDIA GeForce RTX 4070
• DeepStream Version 7.0
• JetPack Version (valid for Jetson only)
• TensorRT Version 8.6.1.6-1+cuda12.0
• NVIDIA GPU Driver Version (valid for GPU only) 535.216.01
• 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)

Im using deepstream analytics Python app with Detectnet_v2 retrained in my custom data. The line crossing is missing many crossing objects. Please, see the result below:

Here are the input video and the configurations:
config_nvdsanalytics.txt (2.7 KB)

3

Thanks for the sharing! noticing some cars are not detected, could you highlight which cars are not coutned by line corssing?

4

@fanzh, thank you for replying!
To be more specific, I have noticed that the application fail to count cars that have slow velocity. The last video that I uploaded, the application fail all count. Unfortunately a can not upload a video with more than 100mbytes.
How can I upload my model (.engine) so that you can reproduce the error?

5

In the first video above, At which second and which cars(you can point the carid) are not counted? you may split the file to multiple files by zip.

6

Here, car ID 7 and 9 are not counted.

7

In the first video, car ID 9 (0:10 second) and car ID 7 (0:15 second) not count.

8

Here is the model that Im using:
resnet18_detector.onnx_b1_gpu0_fp16.zip (19.5 MB)

9

which deepstream sample are you testing or referrng to? what is the whole media pipeline?
noticing you are testng a hight resoluton video, the resnet18_detector.onnx_b1_gpu0_fp16.zip you shared is a TRT engine file, which is is hardware dependent, to reprodcue this issue, could you share the onnx model and configuration file?

10

The problem was the streammux size:
streammux.set_property(‘width’, 856)
streammux.set_property(‘height’, 480).
When I changed to:
streammux.set_property(‘width’, 2720)
streammux.set_property(‘height’, 1530)
The problem was solved. Now all line crossing are counting.