search

triple-Mu / YOLOv8-TensorRT

YOLOv8 using TensorRT accelerate !
MIT License
1.23k stars 210 forks source link

Deepstream compatible yolov8-seg Custom mask and bbox plugin #215

Open neeraj-satyaki opened 2 months ago

neeraj-satyaki commented 2 months ago

Hi triple-Mu, I am really impressed by your work, its a lot of contribution for Yolov8. I would request one more thing,

I am trying to export the model from nms_seg_plugin branch export code for some reason I could not export it. Anyway I was able to do it with yolo cli, it looks like, tensorrt engine has different set of output layers. like User 0 INPUT kFLOAT images 3x640x640 1 OUTPUT kFLOAT output1 32x160x160 2 OUTPUT kFLOAT output0 38x8400 Which is handled using CV2. Could you give an idea, how we can implement this in deepstream custom parser plugin without CV2. Or directly implement it.

triple-Mu commented 1 month ago

Maybe you can get more info in https://github.com/triple-Mu/YOLOv8-TensorRT/blob/main/csrc/segment/normal/include/yolov8-seg.hpp#L229-L347