YOLOv8 - What is the clean and correct way to specify augmentation parameter?

Question

as the title says, how do I set parameters for augmentation while using YOLOv8? I want to use the Python SDK and not the CLI commands.

The mantainer of the repo refer several times to https://docs.ultralytics.com/modes/predict/#arguments but I don't understand where to store these parameters? Also, I found this.

[Comment section from this site](https://i.stack.imgur.com/4qWX2.png)

It would be nice to seperate the settings from data.yaml, but a working solution is the priority

Answer 1

You can store the augmentation parameters directly to the train function like this:

results = model.train(data='coco128.yaml', epochs=300, degrees=0.25, scale=0.3, perspective=0.0001)

List of augmentation parameters: https://docs.ultralytics.com/usage/cfg/#augmentation

Answer 2

In YOLOv8, the augmentation configuration can be found at ultralytics/yolo/cfg/default.yaml. You can change the YAML file directly or set the parameter in model.train() as mentioned above.