Merge Layers using TensorRT- Yolov3

I am trying to create a faster version of Yolov3 using TensorRT. As far as i know, this can be achieved by merging layers. But i don’t have any idea how to achieve it. If there is some reference or any alternate approach, kindly share.

PS: I don’t want to use Deepstream plugin, as i have made changes in the default Yolov3 code for improving accuracy and speed.

I am using
Cuda- 9.0
Python 3.5


Here is our tutorial for YOLO2 and YOLO3 with TensorRT for your reference:

It does use the Deepstream plugin. If you have made custom modifications, then depending if you used supported operations or not (, you may have to impliment custom layers:

using C++:
using Python:

NVIDIA Enterprise Support

Hey @NVES,
Can you please clarify one doubt.

Which part of Tensorrt actually merge the layers. Is it the builder or parser that automatically merge the layers? or we have to manually do it?