Skip to main content

Vision documentation

The Vision service is responsible for image annotations and analyses.

Annotations​

Images uploaded to Animeshon might be automatically analyzed and annotated through an internal AI to extract features and better classify an image. This is especially required to detect explicit content that could be discarded from search results or hidden during navigation depending on the user preferences.

Text Annotations​

Text annotations detect any text within the image and attempt to split such content in paragraphs.

{
  "textAnnotations": [
    {
      "property": {
        "languageCode": "eng",
        "confidence": 0.9
      },
      "boundingBox": {
        "vertices": [
          {"X": 0, "Y": 0},
          {"X": 0, "Y": 10},
          {"X": 10, "Y": 0},
          {"X": 10, "Y": 10},
        ]
      },
      "text": "Test",
      "confidence": 0.8
    }
  ]
}

Label Annotations​

Label annotations detect abstract features of the image such as actions, sentiment, and so on.

{
  "labelAnnotations": [
    {
      "name": "kissing",
      "score": 0.8,
      "topicality": 0.9
    }
  ]
}

Entity Annotations​

Entity annotations detect objects within the image such as body parts, clothes, and so on.

{
  "entityAnnotations": [
    {
      "name": "human-eye",
      "score": 0.8,
      "boundingBox": {
        "vertices": [
          {"X": 0, "Y": 0},
          {"X": 0, "Y": 10},
          {"X": 10, "Y": 0},
          {"X": 10, "Y": 10},
        ]
      },
    }
  ]
}

Graph Knowledge Annotations​

Graph knowledge annotations detect entities that are known by Animeshon such as animes, characters, and so on.

{
  "labelAnnotations": [
    {
      "resource": "animes/12345",
      "score": 0.8,
      "boundingBox": null
    }
  ]
}

WebSearch Annotations​

WebSearch annotations detect potentially associated web resources known by Animeshon such as links to social media.

{
  "webSearchAnnotations": [
    {
      "url": "https://www.pixiv.net/my-link",
      "score": 1.0
    }
  ]
}

SafeSearch Annotations​

SafeSearch annotations detect potentially explict content that should be shown or hidden based on the user preferences.

{
  "safeSearchAnnotation": {
    "Adult":    "POSSIBLE",
    "Racy":     "POSSIBLE",
    "Violence": "POSSIBLE",
    "Medical":  "POSSIBLE",
    "Juvenile": "POSSIBLE"
  }
}