Labels swapped by yolo

[MateuszGalinski]

My issue is as stated in the title, during training my labels got swapped, model correctly finds object but completely misslabels them for example Person is labeled as no-goggles. Dataset was downloaded from roboflow and used as is

this is my data.yaml file

train: C:/Users/LocalAdminUser02/Desktop/PPEDetection/dataset/train/images
val: C:/Users/LocalAdminUser02/Desktop/PPEDetection/dataset/valid/images
test: C:/Users/LocalAdminUser02/Desktop/PPEDetection/dataset/test/images

nc: 7
names: ['Person', 'goggles', 'helmet', 'no-goggles', 'no-helmet', 'no-vest', 'vest']

roboflow:
  workspace: ai-camp-safety-equipment-detection
  project: ppe-detection-using-cv
  version: 3
  license: CC BY 4.0
  url: https://universe.roboflow.com/ai-camp-safety-equipment-detection/ppe-detection-using-cv/dataset/3

example labeled image:

and labels for it:

2 0.3560307692307692 0.19285564853556486 0.23975384615384615 0.3117991631799163
1 0.36552307692307695 0.3049581589958159 0.13053846153846152 0.10712343096234309
6 0.24475384615384616 0.6526150627615063 0.3556692307692308 0.5130334728033472
0 0.35052307692307694 0.47444560669456065 0.5712692307692308 0.8710355648535565

I train model like this results = model.train(data = os.path.join("dataset_goggles", "data.yaml"), epochs = 100) and try to load best.pt best_model = YOLO(os.path.join("runs", "detect", "train", "weights", "best.pt")) and plot images as follows:

for result in results:
    img = result.plot()

img_rgb = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)

plt.imshow(img_rgb)
plt.axis('off')  # Hide axes
plt.show()

and results I get are following

Any suggestions, only thing I found about swapped labels is if you use spaces in classes names but i dont use any?

[glenn-jocher]

@MateuszGalinski it appears that your labels might be mismatched due to an issue with the dataset or the training process. Please ensure that the class order in your data.yaml file matches the order in your dataset's annotations. Additionally, verify that the issue persists with the latest version of the Ultralytics package. If the problem continues, consider re-downloading the dataset from Roboflow to ensure data integrity.

[MateuszGalinski]

It turns out I had a typo in .yaml file and used a dataset from different path. Thank you for your help!

[glenn-jocher]

@MateuszGalinski i'm glad to hear you resolved the issue! If you encounter any more problems or have further questions, feel free to reach out. Happy training!