Fix cutout bug

[Oswells]

Signed-off-by: Junjie Zhang [email protected]

When I use cutout, I find that even if the object is completely covered, its gt_label is still retained on the image.

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Refinement of cutout augmentation in data preprocessing to enhance object detection.

📊 Key Changes

• Imported xywhn2xyxy function within augmentations.py.
• Improved the cutout function by ensuring it considers the IOA (Intersection over Area) of bounding boxes in the image coordinate space.

🎯 Purpose & Impact

• The purpose of these changes is to increase the accuracy of the cutout data augmentation technique used during training.
• By converting labels to the correct coordinate format before calculating IOA, it ensures that the object detection model is trained on better quality data, potentially leading to improved model performance.
• Users can expect their models to handle partial occlusions in a more robust way, which can be particularly beneficial for real-world applications where objects are often partially obscured.

[github-actions]

👋 Hello @Oswells, thank you for submitting a YOLOv5 🚀 PR! To allow your work to be integrated as seamlessly as possible, we advise you to:

• ✅ Verify your PR is up-to-date with ultralytics/yolov5 master branch. If your PR is behind you can update your code by clicking the 'Update branch' button or by running git pull and git merge master locally.
• ✅ Verify all YOLOv5 Continuous Integration (CI) checks are passing.
• ✅ Reduce changes to the absolute minimum required for your bug fix or feature addition. "It is not daily increase but daily decrease, hack away the unessential. The closer to the source, the less wastage there is." — Bruce Lee

[glenn-jocher]

@Oswells thanks for the PR! It looks like you are saying that the current box inputs are in the wrong format. Are you sure this is the case?

[Oswells]

Yes, I think the current box inputs are in the wrong format.
I use this code to check and it works well on yolov5:v6.1, but it may not be compatible with yolov5: v6.2. When I test it on yolov5:v6.2, I use v6.1's plot_images. I find that the latest version still exist this problem.

from utils.datasets import create_dataloader
from utils.general import colorstr,init_seeds
from utils.plots import plot_images
import yaml 

init_seeds(429)
hyp='data/hyps/hyp.scratch-low.yaml'
with open(hyp, errors='ignore') as f:
            hyp = yaml.safe_load(f)
dataloader, dataset = create_dataloader(  "test.txt",
                                            480,
                                            22 // 2,
                                            32,
                                            False,
                                            hyp=hyp,
                                            augment=True,
                                            cache="ram",
                                            rect=False,
                                            rank=-1,
                                            workers=4,
                                            image_weights=False,
                                            quad=False,
                                            prefix=colorstr('train: '),
                                            shuffle=False)
for imgs, targets, paths, _ in dataloader:
    plot_images(imgs,targets)

before：

fixed：

[glenn-jocher]

@Oswells I reviewed. You are absolutely right, the labels are in xywhn so the IoA computation is flawed.

Your fix looks good!

[glenn-jocher]

@Oswells PR is merged. Thank you for your contributions to YOLOv5 🚀 and Vision AI ⭐