Jetson Nano ZED Tensorflow


Has anyone managed to get decent performance (or any) running ssd_mobilenet_v2_coco or similar on stereo footage with the ZED SDK using the Jetson Nano?

I have tried with tf1.13&1.15.0 and Jetpack 4.2 and found the GPU memory capacity prevents it getting much further than a few frames if any.
I optimised ssd_mobilenet_v2_coco in GitHub - NVIDIA-AI-IOT/tf_trt_models: TensorFlow models accelerated with NVIDIA TensorRT as a TRT model in tf 1.15 and Jetpack 4.2. This slightly degraded performance.

Next on the menu is tf1.15.2 with JP4.3.

Interested to find out if it is a lost cause.


Sorry for the late update.

We don’t have too much experience for ZED SDK.
But here are some profiling result for ssd_mobilenet_v2_coco on Nano: