This document contains current restrictions and usage quotas on use of AutoML Vision Object Detection. This page will be updated to reflect any changes to these restrictions and usage quotas.
These quotas apply to each AutoML Vision Object Detection project and are shared across all applications and IP addresses using a given project.
You can edit your quotas up to their maximum values by selecting Edit Quotas from the Quotas page of the Google Cloud Platform Dashboard. Edit the quotas for the Cloud AutoML API. To request an increase above the maximum quota, edit your quota with your requested increase and justification and submit your update. You will be notified when your request has been received. You might be contacted for more information regarding your request. After your request has been reviewed, you will be notified whether it has been approved or denied.
AutoML Vision Object Detection has quotas associated with creating models, human labeling, and making requests to the models.
Modeling quotas
| Activity | Default quota |
|---|---|
| Concurrent model training (creation) requests | 5 |
Request quotas
The current usage quotas for AutoML Vision Object Detection for non-prediction requests is 600 per minute per project. It applies collectively to all deployed models.
Online prediction request quotas
The number of online prediction per minute for AutoML Vision Object Detection are based on how many nodes are deployed. As is explained on the pricing page, many customers find that with one node hour they can serve maximum 1.5 QPS. You can adjust the number of nodes when you deploy your model. However, the total number of nodes across all the deployed models in the project is subject to the quota below.
| Activity | Default quota |
|---|---|
| The number of deployed model nodes | 10 |
Offline batch prediction requests
The current usage quotas for AutoML Vision Object Detection are 5 concurrent batch prediction requests per project. These quotas apply collectively to all deployed models in a project.
Limits
| Type of limit | Value |
|---|---|
| Image file size | Maximum: 30MB |
| Images per dataset | Maximum: 1,000,000 |
| Labels per dataset | Minimum: 1 Maximum: 1,000 |
| Images per label | Minimum: 10 Recommended: 1,000 |
| Batch input CSV file size | Maximum: 100MB |
| Number of images in batch input | Maximum: 300,000 |
Review quotas on Google Cloud console
There are two main ways to view your current quota limits in the Google Cloud console:
- Using the Quotas page, which gives you a list of all your project's quota usage and limits.
- Using the console, which gives you quota information for a particular API, including resource usage over time.
Locate specific operation quotas in the
Quotas page by first selecting
Cloud AutoML API from the Service menu. With Service:
Cloud AutoML API selected you can then select the appropriate Metric.
Examples:
| Operation description | Method name | Metric name in console |
|---|---|---|
| Image classification: Simultaneous model training | projects.locations.models.create |
"Concurrent image classification model creation requests" |
| Image classification: Online prediction | projects.locations.models.predict |
"Online image classification prediction requests per minute" |
| Object detection: Simultaneous model training | projects.locations.models.create |
"Image object detection concurrent model creation requests" |
| Object detection: Simultaneous offline batch prediction | projects.locations.models.batchPredict |
"Image object detection concurrent batch prediction requests" |
Quotas page:
