search

marcoslucianops / DeepStream-Yolo

NVIDIA DeepStream SDK 7.0 / 6.4 / 6.3 / 6.2 / 6.1.1 / 6.1 / 6.0.1 / 6.0 / 5.1 implementation for YOLO models
MIT License
1.38k stars 343 forks source link

Introduction of ONNX, questions Yolo::buildYoloNetwork existence #454

Open jstumpin opened 9 months ago

jstumpin commented 9 months ago

Ever since the Big update commit, we don't have to build the engine from TensorRT API anymore. Empirically (tested on YOLOv5 & YOLOv8), conversion via trtexec yields identical performance & accuracy. Not sure whether or not Yolo::buildYoloNetwork is a necessary step for DeepStream, but for my use case (non-DeepStream), it offers me the option to build the engine offline and simplify the codebase.

Big update demands big appreciation! THANKS!

marcoslucianops commented 9 months ago

The Yolo::buildYoloNetwork is for Darknet models (YOLOv2, YOLOv3 and YOLOv4).

jstumpin commented 9 months ago

My bad, that is true, although one could convert those into ONNX as well. In fact, Yolo::createEngine is also unneeded with the out-of-source creation via trtexec.