What Does It Mean If a Dataset Contains Errors and Warnings After Evaluation?

What does a "warning" in a dataset evaluation mean?

When you evaluate a dataset, warnings (highlighted in yellow) may appear if certain images, or "outliers," differ significantly from others in the same class. These outliers might affect the training of an AI model, though not always. It's a good idea to review these images and consider excluding them from training if necessary.

Common Warning Examples:

• Sharpness
  If an image is blurry, you'll see a yellow "Blurry Face(s)" warning.

• Outliers
  If an image looks very different from others in the same class, you may receive a red "Outlier" error.

• Multiple Faces
  Images showing more than one face trigger a red "Multiple Faces" error.

• Head Pose
  If the person's head is tilted too far from the camera, a red "Head Pose" error appears, indicating it's not suitable for training.

What makes a dataset "good"?

A dataset is considered "good" (marked in green) when it contains no errors or warnings, making it ideal for AI model training.

Can I train an AI model even if there are warnings or errors?

Yes, you can still train an AI model despite warnings or errors in the dataset.

However, it’s recommended to address these issues, especially errors, to improve the model's performance. Warnings should be reviewed carefully to decide whether the affected images should be excluded from training.