Jetson nano running both the DS resnet10 and the TLT resnet10 comparison

Accelerated Computing Intelligent Video Analytics TAO Toolkit
1

Here is a video comparing the performance of the DS Resnet10.caffemodel and the TLT Resnet10_detector.trt running on the nano.

With some tweaking, headbanging and patience the TLT Resnet10_detector.trt can perform equal to the DS Resnet10.caffemodel

2

Thanks adventuredaisy!!
Appreciate your sharing and contribution.
I see in your video, you were training KITTI dataset and prune it with “pth” 0.01.
Re-train tlt model, export etlt model, generate trt int8 engine and export it into Deepstream.
Pretty good!

3

That‘s amazing,I test tlt resnet10 customer data(544*960 size images), only has 20fps on jetson nx