Oracle Cloud Infrastructure Documentation

Pretrained Models

Vision has four types of pretrained image analysis model.

The pretrained models are:

Object Detection

Object detection is used to find and identity objects within an image. For example, if you have an image of a living room, Vision find the objects there, such as a chair, a sofa, and a TV. It then provides bounding boxes for each of the objects and identifies them.

Vision provides a confidence score for each object identified. The confidence score is a decimal number. Scores closer to 1 indicate a higher confidence in the objects classification, while lower scores indicate a lower confidence score. The range of the confidence score for each label is from 0 to 1.

Supported features are:
  • Labels
  • Confidence score
  • Object-bounding polygons
  • Single requests
  • Batch requests
Object Detection Example

An example of Object Detection use in Vision.

Input image
Figure 1. Input image for Object Detection
Picture of a car driving on a road, with Oracle written on it. To one side is a bus, to the other a taxi.
API Request:
{ "analyzeImageDetails":
 { "compartmentId": "",
   "image": 
           { "namespaceName": "",
             "bucketName": "",
             "objectName": "",
             "source": "OBJECT_STORAGE" },
   "features": [ { "featureType": "OBJECT_DETECTION", "maxResults": 50 } ] } }
Output:
Figure 2. Output image for Object Detection
The same image as the input, but now with bounding boxes drawn around all items of interest.
API Response: 
{ "imageObjects": 
 [ { "name": "Bus",
     "confidence": 0.98872757,
     "boundingPolygon": 
                       { "normalizedVertices": 
                                              [ { "x": 0.232, 
                                                  "y": 0.16114457831325302 },
                                                { "x": 0.407,
                                                  "y": 0.16114457831325302 }, 
                                                { "x": 0.407,
                                                  "y": 0.36596385542168675 },
                                                { "x": 0.232,
                                                  "y": 0.36596385542168675 } ]
                       } },
   }, ... }

Image Classification

Image classification can be used to identify scene-based features and objects in an image. You can have one classification or many classifications, depending on the use case and the number of items in an image. For example, if you have an image of a person running, Vision identifies the person, the clothing, and the footwear.

Vision provides a confidence score for each label. The confidence score is a decimal number. Scores closer to 1 indicate a higher confidence in the label, while lower scores indicate lower confidence score. The range of the confidence score for each label is from 0 to 1.

Supported features are:
  • Labels
  • Confidence score
  • Ontology classes
  • Single requests
  • Batch requests
Image Classification Example

An example of Image Classification use in Vision.

Input image
Figure 3. Input image for Image Classification
Picture of electricity pylons crossing a road.
API Request:
{ "analyzeImageDetails":
 { "compartmentId": "",
   "image":
           { "namespaceName": "",
             "bucketName": "",
             "objectName": "",
             "source": "OBJECT_STORAGE" },
             "features":
                        [ { "featureType": "IMAGE_CLASSIFICATION",
                            "maxResults": 5 } ]
 } 
}
Output:
API response: 
{ "labels":
           [ { "name": "Overhead power line",
               "confidence": 0.99315816 },
             { "name": "Transmission tower",
               "confidence": 0.9927904 },
             { "name": "Plant", "confidence": 0.9924676 },
             { "name": "Sky", "confidence": 0.9924451 },
             { "name": "Line", "confidence": 0.9912027 } ] ...

Face Detection

Vision can detect and recognize faces in an image.

Face detection lets you pass an image or a batch of images to Vision to detect the following using a pretrained model:

  • The existence of faces in each image.
  • The location of faces in each image.
  • Face landmarks for each face.
  • Visual quality of each face.

No data science experience is required to use this pretrained model.

Face Example

An example of face detection in Vision.

Input image
Figure 4. Input image for face detection
Picture of a motorbike next to a car.
API Request:
{
  "compartmentId": "ocid1.compartment.oc1..aaaaaaaau3mwjanch4k54g45rizeqy52jcaxmiu4ii3kwy7hvn6pncs6yyba",
  "image": {
    "namespaceName": "axwlrwe7tbir",
    "bucketName": "demo_examples",
    "objectName": "FaceDetection/FaceDetection1.jpeg",
    "source": "OBJECT_STORAGE"
  },
  "features": [
    {
      "featureType": "FACE_DETECTION",
      "maxResults": 50,
      "shouldReturnLandmarks": true
    }
  ]
}
Output:
Figure 5. Output Image for face detection
The texts in the input image are surrounded by bounding boxes.
API response: 
{
  "ontologyClasses": [],
  "detectedFaces": [
    {
      "confidence": 0.9838427,
      "boundingPolygon": {
        "normalizedVertices": [
          {
            "x": 0.48696465492248536,
            "y": 0.2889890061576746
          },
          {
            "x": 0.6339863777160645,
            "y": 0.2889890061576746
          },
          {
            "x": 0.6339863777160645,
            "y": 0.586297366400352
          },
          {
            "x": 0.48696465492248536,
            "y": 0.586297366400352
          }
        ]
      },
      "qualityScore": 0.9043028,
      "landmarks": [
        {
          "type": "LEFT_EYE",
          "x": 0.5203125,
          "y": 0.41114983
        },
        {
          "type": "RIGHT_EYE",
          "x": 0.590625,
          "y": 0.41231126
        },
        {
          "type": "NOSE_TIP",
          "x": 0.553125,
          "y": 0.4715447
        },
        {
          "type": "LEFT_EDGE_OF_MOUTH",
          "x": 0.5210937,
          "y": 0.5005807
        },
        {
          "type": "RIGHT_EDGE_OF_MOUTH",
          "x": 0.5914062,
          "y": 0.5017422
        }
      ]
    },
    {
      "confidence": 0.9775677,
      "boundingPolygon": {
        "normalizedVertices": [
          {
            "x": 0.7882407665252685,
            "y": 0.26365977075734065
          },
          {
            "x": 0.9403343200683594,
            "y": 0.26365977075734065
          },
          {
            "x": 0.9403343200683594,
            "y": 0.5528718281567582
          },
          {
            "x": 0.7882407665252685,
            "y": 0.5528718281567582
          }
        ]
      },
      "qualityScore": 0.786416,
      "landmarks": [
        {
          "type": "LEFT_EYE",
          "x": 0.81328124,
          "y": 0.37514517
        },
        {
          "type": "RIGHT_EYE",
          "x": 0.88125,
          "y": 0.39140534
        },
        {
          "type": "NOSE_TIP",
          "x": 0.8296875,
          "y": 0.44134727
        },
        {
          "type": "LEFT_EDGE_OF_MOUTH",
          "x": 0.8078125,
          "y": 0.46689895
        },
        {
          "type": "RIGHT_EDGE_OF_MOUTH",
          "x": 0.8726562,
          "y": 0.48083624
        }
      ]
    },
    {
      "confidence": 0.97464997,
      "boundingPolygon": {
        "normalizedVertices": [
          {
            "x": 0.038544440269470216,
            "y": 0.2764744597998784
          },
          {
            "x": 0.17794162034988403,
            "y": 0.2764744597998784
          },
          {
            "x": 0.17794162034988403,
            "y": 0.560027438173726
          },
          {
            "x": 0.038544440269470216,
            "y": 0.560027438173726
          }
        ]
      },
      "qualityScore": 0.8527186,
      "landmarks": [
        {
          "type": "LEFT_EYE",
          "x": 0.08984375,
          "y": 0.3809524
        },
        {
          "type": "RIGHT_EYE",
          "x": 0.15234375,
          "y": 0.39140534
        },
        {
          "type": "NOSE_TIP",
          "x": 0.12421875,
          "y": 0.44599304
        },
        {
          "type": "LEFT_EDGE_OF_MOUTH",
          "x": 0.07734375,
          "y": 0.46689895
        },
        {
          "type": "RIGHT_EDGE_OF_MOUTH",
          "x": 0.14375,
          "y": 0.47619048
        }
      ]
    },
    {
      "confidence": 0.96874785,
      "boundingPolygon": {
        "normalizedVertices": [
          {
            "x": 0.2698225736618042,
            "y": 0.24420403492713777
          },
          {
            "x": 0.38425185680389407,
            "y": 0.24420403492713777
          },
          {
            "x": 0.38425185680389407,
            "y": 0.4686152760575457
          },
          {
            "x": 0.2698225736618042,
            "y": 0.4686152760575457
          }
        ]
      },
      "qualityScore": 0.8934359,
      "landmarks": [
        {
          "type": "LEFT_EYE",
          "x": 0.29453126,
          "y": 0.3240418
        },
        {
          "type": "RIGHT_EYE",
          "x": 0.3484375,
          "y": 0.33681765
        },
        {
          "type": "NOSE_TIP",
          "x": 0.31328124,
          "y": 0.3821138
        },
        {
          "type": "LEFT_EDGE_OF_MOUTH",
          "x": 0.2890625,
          "y": 0.39372823
        },
        {
          "type": "RIGHT_EDGE_OF_MOUTH",
          "x": 0.3453125,
          "y": 0.40301976
        }
      ]
    },
    {
      "confidence": 0.95825064,
      "boundingPolygon": {
        "normalizedVertices": [
          {
            "x": 0.6876011371612549,
            "y": 0.10002164585942037
          },
          {
            "x": 0.8045546531677246,
            "y": 0.10002164585942037
          },
          {
            "x": 0.8045546531677246,
            "y": 0.3600864033804261
          },
          {
            "x": 0.6876011371612549,
            "y": 0.3600864033804261
          }
        ]
      },
      "qualityScore": 0.9237982,
      "landmarks": [
        {
          "type": "LEFT_EYE",
          "x": 0.7171875,
          "y": 0.19976771
        },
        {
          "type": "RIGHT_EYE",
          "x": 0.7703125,
          "y": 0.21254355
        },
        {
          "type": "NOSE_TIP",
          "x": 0.7367188,
          "y": 0.2601626
        },
        {
          "type": "LEFT_EDGE_OF_MOUTH",
          "x": 0.7085937,
          "y": 0.2752613
        },
        {
          "type": "RIGHT_EDGE_OF_MOUTH",
          "x": 0.76640624,
          "y": 0.2857143
        }
      ]
    }
  ],
  "faceDetectionModelVersion": "1.0.27",
  "errors": []
}

Optical Character Recognition (OCR)

Vision can detect and recognize text in a document.

Language classification identifies the language of a document, then OCR draws bounding boxes around the printed or hand-written text it locates in an image, and digitizes the text. For example, if you have an image of a stop sign, Vision locates the text in that image and extracts the text STOP. It provides bounding boxes for the identified text.

Vision provides a confidence score for each text grouping. The confidence score is a decimal number. Scores closer to 1 indicate a higher confidence in the extracted text, while lower scores indicate lower confidence score. The range of the confidence score for each label is from 0 to 1.

Text Detection can be used with Document AI or Image Analysis models.

OCR support is limited to English. If you know the text in your images is in English, set the language to Eng.

Supported features are:
  • Word extraction
  • Text line extraction
  • Confidence score
  • Boundling polygons
  • Single request
  • Batch request
OCR Example

An example of OCR use in Vision.

Input image
Figure 6. Input image for OCR
Picture of a motorbike next to a car.
API Request:
{ "analyzeImageDetails":
 { "compartmentId": "",
   "image":
           { "namespaceName": "",
             "bucketName": "",
             "objectName": "",
             "source": "OBJECT_STORAGE" },
   "features":
              [ { "featureType": "TEXT_DETECTION" } ]
 }
}
Output:
Figure 7. Output Image for OCR
The texts in the input image are surrounded by bounding boxes.
API response: 
...
 { "text": "585-XRP",
   "confidence": 0.9905539,
   "boundingPolygon":
                     { "normalizedVertices":
                                            [ { "x": 0.466,
                                                "y": 0.7349397590361446 },
                                              { "x": 0.552,
                                                "y": 0.7319277108433735 },
                                              { "x": 0.553,
                                                "y": 0.7831325301204819 },
                                              { "x": 0.467,
                                                "y": 0.7876506024096386 } ]
                     }
 } 
...

Using the Pretrained Image Analysis Models

Vision provides pretrained models for customers to extract insights about their images without needing Data Scientists.

You need the following before using a pretrained model:

  • A paid tenancy account in Oracle Cloud Infrastructure.

  • Familiarity with Oracle Cloud Infrastructure Object Storage.

You can call the pretrained Image Analysis models as a batch request using Rest APIs, SDK, or CLI. You can call the pretrained Image Analysis models as a single request using the Console, Rest APIs, SDK, or CLI.

See the Limits section for information on what is allowed in batch requests.