Search code examples
google-cloud-platformobject-detectiongoogle-cloud-vision

Can't train Google Vision, because Validation errors

I have a problem with Google Cloud Vision. I added 16 pictures and I added bounding boxes. But I have an error for Validation, and I don't understand why ?!

I have export my CSV for add VALIDATION, but when I upload this new CSV, nothing changes... I have already this error.

And with that, I can't start training.

enter image description here

Solution

  • Google recommends about 1000 training images per label. The minimum per label is 10, or 50 for advanced models. In general it takes more examples per label to train models with multiple labels per image. The AutoML Vision uses 80% of your content documents for training, 10% for validating, and 10% for testing. The maximum size of a test dataset is 50,000 images, even if 10% of the total dataset exceeds that maximum.

    As mentioned in the error message, each label should have at least 10 bounding boxes. You must also have at least 8,1,1 bounding boxes each assigned to your Train, Validation and Test sets. It is recommended to have a larger image set than the minimum required when relying on the auto assignment feature.

    Also check this documentation on preparing training data that explains the minimum requirements that have to be satisfied in order to train your data.