Announcing OCI Vision service

We’re excited to announce the general availability of Oracle Cloud Infrastructure (OCI) Vision, a computer vision service that allows customers to uncover insights in unstructured images powered by deep learning models.

What is OCI Vision?

OCI Vision is a serverless, cloud native service that provides deep learning-based, prebuilt, and custom computer vision models over REST APIs. OCI Vision helps you identify and locate objects, extract text, and identify tables, document types, and key-value pairs from business documents like receipts. No data science experience is required to use the prebuilt or custom features of OCI Vision.

You can access the service through the Oracle Cloud Console, OCI software developer kits (SDKs) in Python and Java, or the OCI CLI.

OCI Vision key capabilities

Image Analysis models available in the analyzeImage API help you understand images like photos of streets, retail goods, and transmission towers.

Image Analysis includes the following key features:

◉ Image classification: Assigns labels to the image based on the overall scene, such as "Sky,” “moisture,” and “textile”

◉ Object detection: Locates and identifies objects within an image, such as a bus, box, or person

◉ Text recognition and optical character recognition (OCR): Locates and digitizes text information from images, such as “Stop” from a stop sign or “XY3497” from a license plate

◉ Async and batch support on all image analysis APIs

Document AI models available in the analyzeDocument API help you understand document-based images like receipts, invoices, and contracts.

Document AI includes the following key features:

◉ Text recognition and OCR: Locates and digitizes text information from images at the word or line level

◉ Key-value extraction: Extract a predefined list of key-value pair information from receipts, such as fieldLabel: “TransactionDate” and fieldValue: “01/11/2022.”

◉ Table extraction: Extracts content in tabular format, maintaining row/column relationships of cells, such as cell text: “2,098,221” rowIndex: 14 columnIndex: 2.

◉ Document classification: Classifies documents into different types based on visual appearance, high-

level features, and extracted keywords, such as invoice, receipt, and resume

◉ Async and batch support on all document analysis APIs

Custom models

Customers without data science expertise can easily use OCI Vision to fit their industry or customer-specific use cases. With pretrained models for out-of-the-box use, OCI Vision also supports creating custom image classification and object detection models. Training and underlying model infrastructure are all managed through OCI Vision.

Using custom model features

To train a custom model using OCI Vision service, start with a labeled dataset. You can easily label raw images with OCI Data Labeling service.

After you select your model type and dataset, name your model and select a training duration. The default is “Recommended.”

After you kick off training a new model, model training progress, logs, and final quality metrics are available on the Model Details page. You also have an "Analyze" option to test a newly trained model on new images.

To call your custom vision model, include the model OCID as part of the modelID field in your input request. The following example shows a JSON request to call a custom image classification model:

{

"analyzeImageDetails": {

"compartmentId": "ocid1.tenancy.oc1..xxxx",

"image": {

"source": "INLINE",

"data": "......"

},

"features": [

{

"modelId": "ocid1.aivisionmodel.oc1.iad.amaaaaaapheaxxxxxxxxxxx",

"featureType": "IMAGE_CLASSIFICATION",

"maxResults": 5

}

]

}

}

Computer vision use cases

Use cases of computer vision exist in many industry verticals, including financial services, manufacturing, transportation, and retail.

◉ Automate back-office tasks: Classify documents, detect tables, and extract required information from documents like receipts to automate business workflows including employee expense reporting and reimbursement.

◉ Digital asset management: Enrich image-based files with metadata including document type, text, and objects for better indexing and retrieval in a digital asset management system or larger data warehouse.

◉ Detect visual anomalies: Classify products or equipment as standard or defective based on visual appearance like discoloration, tear, rust, deformity, or breaks. Automate the detection of defective materials to flag the need for repairs.

Computer vision customer momentum

As part of building OCI Vision, we partnered closely with customers across financial services, media, and large-scale Oracle software-as-a-service (SaaS) teams:

2RP is streamlining manual document scanning processes in the banking sector. "We were impressed with Oracle's performance on both good and less-than-ideal quality images, the latter being a common complaint in financial onboarding scenarios.” - Nicolas Borges, Product Owner.

SailGP is tagging objects in their race images. “We’re excited about Oracle Cloud Infrastructure Vision because of its ability to quickly process large amounts of visual data and the potential to increase the productivity of our teams.” - Aleksandar Kocic, Solution Architect.

Oracle Analytics Cloud is enabling business analysts the power to analyze visual data with OCI Vision service. “In 2022, Oracle Analytics is integrating with Oracle Cloud Infrastructure Artificial Intelligence services. This service detects objects and text and classifies them, making it possible for you to analyze images directly inside Oracle Analytics.”

Source: oracle.com