I am working on using the Jetson Nano to perform the function of human detection. I have tried all of the pre-trained models provided in the Jetson Inference Github, and unfortunately, all are far from accurate. The number of false positives is way too many.
I have tried training my own model using the Re-train SSD guide, but I don’t know if I have the amount of data needed to create an accurate model.
Faster-RCNN-Inception-V2 was incredibly accurate; however, incredibly slow and does not run in TensorRT. Could someone please advise on any insight they have on how to get a quick and reliable human detection model up and running on Jetson Nano?