Does Yolov8 includes input normalization pipeline?

Discussion
, , , ,
1

Hello, was wondering if I need to include my own input normalization pipeline to Yolov8. I code in C++ and can’t find any information which states if Yolov8 includes input normalization.

I have exported Yolov8 from .pt to .engine (TensoRT) using Ultralytics CLI.

Also if there are any best practice guideline documentation for Yolov8 inference which can be shared or referenced, would be great.

Thanks for any advice.

2

The model doesn’t. You have to manually preprocess. There are C++ examples in the repository.

2 Likes
3

Hello Toxite,

Thank you for replying, I am testing the examples and things makes more sense.

4

YOLOv8 includes an input normalization pipeline, which scales input pixel values to a 0-1 range by dividing by 255. This is handled internally during training and inference. Since you’ve exported your model to TensorRT, ensure your input preprocessing matches this normalization. For best practices and guidelines, you can refer to the Ultralytics Docs. Let me know if you need further clarification!

1 Like