How to apply k-fold cross-validation in yolov5？

[Chenzkun]

hello@glenn-jocher，
I want to use k-fold cross-validation in yolov5, but I don't know how to do it

[spacewalk01]

https://www.kaggle.com/code/ayuraj/train-yolov5-cross-validation-ensemble-w-b

[spacewalk01]

Split your data into k-folds as follows and train a model on each dataset

# Create .yaml file 
import yaml

for fold in range(NUM_FOLD):
    data_yaml = dict(
        train = f'../dataset_folds_{fold}/images/train',
        val = f'../dataset_folds_{fold}/images/valid',
        nc = 2,
        names = ['none', 'opacity']
    )

    # Note that I am creating the file in the yolov5/data/ directory.
    with open(f'yolov5/data/data_fold_{fold}.yaml', 'w') as outfile:
        yaml.dump(data_yaml, outfile, default_flow_style=True)
    
%cat yolov5/data/data_fold_0.yaml

[Chenzkun]

Is this to create a new yaml file to replace coco128.yaml in the source code?

[spacewalk01]

Yes

[Chenzkun]

But you've only created a few yaml files, it doesn't say how to train them

[TonSarun]

https://docs.ultralytics.com/guides/kfold-cross-validation/#setup