AutoML Vision pricing depends on what feature you are using: image classification, object detection, or AutoML Vision Edge.
AutoML Vision enables you to train custom machine learning models to classify images into a custom set of categories.
Prices for usage of AutoML Vision are computed based on the amount of training required (in hours), the amount of human labeling you request, and how many images you send for prediction.
The accuracy of your model generally depends on how long you allow it to train and the quality of your training dataset. You will pay only for the compute hours used.
Prices are listed in US Dollars (USD).
Image classification training
The cost for AutoML Vision model training depends on the number of compute hours used to train the model. The accuracy of your model generally depends on how long you allow it to train and the quality of your training dataset. You pay only for the compute hours used; if you cancel training, you only pay for the compute hours used.
A compute hour represents internal compute usage, and therefore does not exactly match an actual hour on the clock.
You receive one hour of free training per model for up to 10 models each month. Subsequent training hours are USD$20 per hour. If you resume training of an existing model, the training costs USD$20 per hour.
Many customers find that one hour is sufficient to build an experimental model and use additional training hours to increase accuracy to a production level.
Image classification prediction
Your usage of AutoML Vision is calculated in terms of how many images you send to AutoML Vision for prediction. The pricing is tiered: the first 1000 images sent each month are free, images 1001 to 5,000,000 are priced as marked.
|Training||1 node hour per account for the first 10 models each month||Subsequent training hours are USD$20.00 per hour|
|Prediction||First 1,000 images are free||For 1,001–5,000,000 images*, the price is $3 per 1,000 images|
* For 5,000,001 or more images contact Sales for volume discounts.
AutoML Vision Object Detection enables you to train custom object detection models to localize a custom set of objects in your images.
Prices for AutoML Vision Object Detection are based on resource usage, for both training and classification online prediction.
Prices are listed in US Dollars (USD).
Object detection training costs
The cost for training a AutoML Vision Object Detection model is $3.15 per node hour.
For each unit of time, we use 9 nodes in parallel, where each node is equivalent to a n1-standard-8 machine with an attached NVIDIA® Tesla® V100 GPU. See Table below*.
The time required to train your model depends on the size and complexity of your training data. Many customers find that 40 node hours (approximately 5 "wall clock" hours) are sufficient to build a model with 5,000 labeled images or less.
You pay only for the compute hours used; if training fails for any reason other than a user-initiated cancellation, you will not be billed for the time. You will be charged for training time if you cancel the operation.
Object detection deployment and prediction costs
Models must be deployed before they can provide online predictions. You pay per node deployed, even if no prediction is made. There is no additional charge for each prediction served. Note that GPUs remain allocated for your model so that your predictions are not delayed by startup latency.
The cost for deployment and prediction is $1.82 per node hour. For each unit of time, we use 1 node equivalent to a n1-standard-4 machine with an NVIDIA® P100 GPU. See Table below**.
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.
|Training||First 40 node hours are free||*USD$3.15 per node hour|
|Deployment and Prediction||First 40 node hours are free||**USD$1.82 per node hour|
AutoML Vision Edge
The cost for training a AutoML Vision Edge model for image classification is $4.95 per node hour.
For each unit of time, we use 1 node, where the node is equivalent to a Cloud TPU v2 machine.
Free Tier for training is 3 node hours per month/project. Many customers find that 3 node hours is sufficient to build a model with 5k labeled images or less.
You pay only for the compute hours used; if training fails for any reason other than a user-initiated cancellation, you will not be billed for the time. You will be charged for training time if you cancel the operation. Trained models can be exported and downloaded for free.
|AutoML Vision Edge||Free||Paid|
|Training||15 node hours per account (lifetime)*||Subsequent training hours are USD$4.95 per hour|
|Exporting models to edge devices||Free||Free|
* Effective May 7, 2019
Human labeling costs
You can leverage a human labeling service to label your images. The human labeling service is limited to 5000 images per task and is priced as marked.
|Service||Number of raters||Cost|
|Free tier (first 100 images per month)||3||Free|
|Basic tier||1||USD$35 per 1000 images|
|Basic tier||3||USD$100 per 1000 images|
|Premium tier||Contact a partner for a quote|
For the Free tier and Basic tiers, your dataset can have up to 20 labels, and each image can take up to 10 seconds to label.
Google Cloud Platform costs
Since you store images to be analyzed in Google Cloud Storage, and may use other Google Cloud Platform resources in tandem with the AutoML Vision, such as Google App Engine instances, then you will also be billed for the use of those services. See the Google Cloud Platform Pricing Calculator to determine other costs based on current rates.
To view your current billing status in the Cloud Console, including usage and your current bill, see the Billing page. For more details about managing your account, see the Cloud Billing Documentation or Billing and Payments Support.