Fps downgrade during deepstream inference

• Hardware Platform Jetson Nano
• DeepStream Version 5.1
• JetPack Version 4.5
• TensorRT Version 7.1.3
• Issue Type fps downgrade

Hi, I’ve been experimenting several fps downgrades during real-time image classification. I’ve created an engine for my DeepStream app and I made some tests the last few weeks and the average frame rate was 30 fps. Now I’ve added more classes (a total of 97) and I’ve been having this downgrade issues, sometimes 20 fps, sometimes 1fps…

I’ve trained an EfficientNetB1 model with Nvidia TAO toolkit and then I’ve generated the engines using DeepStream app.

Do you know if that downgrades are being caused by the increase of the number of classes?

Please upgrade to Deepstream 6.0 and check again.

That’s because of a known bug or just to try if it works?

We suggest all user work on latest version DS 6.0 and try if still meet same issue.