FPS by image recognition model in Jetson TX2

Robotics & Edge Computing Jetson & Embedded Systems Jetson TX2
1

I want to know FPS by image recognition model in Jetson TX2. Because I want to use FPS best among many models. I tried to cdo jetson-inference and I couldn’t measure the FPS accurately.

  1. I want to see that the FPS by model is accurately quantified.
  2. There are model specific FPSs executed by jetson no, but not image recognition. Can I guess image recognition FPS in jetson with this?
    Thanks.
2

Hi,

Here is the classifier benchmark result for TX2.
https://github.com/NVIDIA-AI-IOT/tf_to_trt_image_classification#models

This table is generated several years ago.
It’s expected that you will get a higher performance with our latest software.

Thanks.

3

Thank you. What does “TX2/Half”, “TX2/Float” mean at that table?

4

Hi,

Sorry for the late update.

TX2/Half and TX2/Float indicate the different precision mode.
You can find more information in this page:
https://devblogs.nvidia.com/tensorrt-3-faster-tensorflow-inference/

Optimization 2: FP16 and INT8 Precision Calibration

Thanks.