Closed OctaM closed 1 year ago
Because it's easier to use any model in the DeepStream. It has a native ONNX parser, so I don't need to create the conversor layer by layer.
Cool, thanks for your answer!
I forgot to say, but the speed and accuracy between both methods are equal.
It's not an actual issue. But i'm just curious why it was changed to .onnx from .wts & .cfg? Is it a performance improvement?