It would be great to have some benchmarking information on good moedel for Face recognition/Comparision done on the Jetson Nano platform.
If Deepstream is used for live video analytics, what are all settings in gstreamer and other pipelines need to be configured ?
We don’t have a face specified benchmark for Nano but you can find some general score here:
For Deepstream, please find the detail information in our document:
Thanks for the benchmark details. Anyways the procedure given seems using the TensorRT as follows.
Is it possible to run the same using opencv. Would like to understand the type of port given to opencv to load CNN models.
The benchmark only calculates the inference time of TensorRT.
The input image is random data rather than an decoded image.
To link OpenCV for inference, you can just fill the buffer from OpenCV image (ex.cudamemcpy).
The type should be float32 and the format is RGB or BGR depends on the model ifself.