Quotas in Amazon Lookout for Vision

The following tables describe the current quotas within Amazon Lookout for Vision. For information about quotas that can be changed, see AWS service quotas.

Model quotas

The following quotas apply to the testing, training, and functionality of a model.

Resource	Quota
Supported file format	PNG and JPEG image formats
Minimum image dimension of image file in an Amazon S3 bucket	64 pixels x 64 pixels
Maximum image dimension of image file in an Amazon S3 bucket	4096 pixels X 4096 pixels is the maximum. Smaller dimensions are able to upload faster.
Differing image dimensions of image files used in a project	All images in the dataset must have the same dimensions
Maximum file size for an image in an Amazon S3 bucket	8 MB
Lack of labels	Images must be labeled as normal or anomaly before training. Images without labels are ignored during training.
Minimum number of images labeled normal in training dataset	10 for a project with separate training and test datasets. 20 for project with a single dataset.
Minimum number of images labeled anomaly in a training dataset	0 for a project with separate training and test datasets. 10 for a project with a single dataset.
Maximum number of images in classification training dataset	16,000
Maximum number of images in a classification test dataset	4,000
Minimum number of images labeled normal in test dataset	10
Minimum number of images labeled anomaly in test dataset	10
Maximum number of images in an anomaly localization training dataset	8000
Maximum number of images in an anomaly localization test dataset	800
Maximum number of images in trial detection dataset	2,000
Maximum dataset manifest file size	1 GB
Maximum number of training datasets in a model	1
Maximum training time	24 hours
Maximum testing time	24 hours
Maximum number of anomaly labels in a project	100
Maximum number of anomaly labels on a mask image	20
Minimum number of images for an anomaly label. To count, the image must contain only one type of anomaly label.	20 for a single dataset project. 10 for each dataset in a project with separate training and test datasets.