Input size config with performances

Please provide complete information as applicable to your setup.

• Hardware Platform (Jetson / GPU)
Xavier NX
• 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)

Hello !

I have a question concerning inference with a tensorrt optimized network.

I don’t understand how reducing the size of the input improves performance:

I trained a detecnet with tao toolkit on 1280x1280 images, I can run it at 80 FPS with deepstream. When I reduce the input resolution of the network to 640x640 I run at 200 FPS. When I increase the input size to 2000x2000 I drop to 20 FPS.

However, I don’t understand how Tensorrt can allow such a performance gain: the network is trained on a fixed image size, does Tensorrt add a scaling layer?

If it is not the case, how are the weights of the network adjusted to proceed to inference on images with a different resolution than the training resolution?

If it is the case, how to explain the performance gain, for the different sizes?

Thanks for your answer.


This looks like a Deepstream related issue. We will move this post to the Deepstream forum.


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. The nvstreammux, nvinfer, nvvideoconvert plugins may do the scaling. Could you show your pipeline and config files?

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