Examples

aion

• Capability Contract: Concepts
  Concepts found in an interview about diabetes. A concept document contains objects and series objects of type ‘concept’. This document contains a summary of all concepts found in a transcription, as well as a series with the concepts found at each time-stamp. This example uses eContext library of concepts, but there are many other libraries as well, like IAB’s Content Taxonomy or IPTC’s Media Topics.
• Capability Contract: Entity
  Entities found in a text message. An entity document contains objects of type ‘namedEntity’. Entity extraction engines classify named entities, located in unstructured text, into predefined categories such as People, Organizations, or Locations.
• Capability Contract: Keyword
  Keywwords and descriptions extracted from a description of a person. A keyword document contains objects of type ‘keyword’. Keyword extraction engines find and extract significant words or phrases from a text, which can be used for indexing, searching, or summarizing content.
• Capability Contract: Language
  The language detected in a document or media file. The only required information is the ‘language’ at the root, which must be an ISO-639-1 language code. This document also includes an optional list of other languages checked, along with their confidence scores.
• Capability Contract: Media-translated
  Translations of full media files. When media is translated, the translation may be attached to TDOs (temporal data objects) as asset documents. The engine stores references to these translation assets in a ‘media’ list, which identifies the language of each document. In this example, literal translations are also stored as optional ‘text’ objects, which makes this document also fulfill the Text Capability Contract.
• Capability Contract: Objects
  Objects identified in a image. An object document only contains objects of type ‘object’. Object detection engines identify various objects in images or video and label their position and attributes. Object detection engines are the most common engine types, and the ‘object’ contract is the most general contract for storing any generic object. There are also more specific object types such as ‘face’, ‘motorVehicle’, ‘logo’, ‘barcode’ that may be more appropriate for specific use cases.
• Capability Contract: Sentiment (emotions)
  Detecting emotions on a face. Emotion analysis engines analyze context or images to determine the emotions portrayed. In this document, a single face is analyzed and the confidence levels for each emotion are provided.
• Capability Contract: Sentiment (sentiments)
  Sentiment detected in a brief review. Sentiment analysis engines analyze text to determine the emotional tone behind it, classifying it as positive, negative, or neutral. Multiple sentiments may be detected in the same phrase or document with varying confidence levels.
• Capability Contract: Summary
  Summary of textual documents. This example contains a summary of a text discussing the development of spider silk in an industrial context.
• Capability Contract: Text (OCR)
  Text read from an image. This example demonstrates the output of OCR on a low-resolution news video.
• Capability Contract: Text (text)
  Text extracted or generated from a document or media file. There are many engine types that generate ‘text’ objects: translations, OCR, and text extraction (there is a separate contract for transcriptions, though). This document contains text translated to Welsh, but any engine that extracts or generates text from an input file can use this contract.
• Capability Contract: Transcript
  Transcription of audio documents. This example contains a transcript of a video about spider silk. Transcript contracts may also be used for translations of spoken language as well.
• Empty Object
  An empty object is a valid AION file
• Object Contract: Concept
  Concept objects may be for specific text phrases, or a list of concepts found across the entire document. Concept objects MUST include ‘text’ and/or ‘objectCategory’ properties.
• Object Contract: Face
  Face object with many supporting fields. Objects of type ‘face’ MUST include a ‘boundingPoly’, SHOULD include ‘label’, ‘confidence’, and MAY include select other fields related to facial features.
• Object Contract: Fingerprint (tracking)
  Fingerprint object used to identify a unique object (person, vehicle, etc.) in a video stream. This fingerprint may also be used to track the object between different videos. These fingerprints MUST include a 256-, 512-, or 1024-dimensional fingerprintVector, referenceId, label, and tags, and MAY include some other properties.
• Object Contract: Fingerprint (watermark)
  Fingerprint object used to identify unique audio or video content. A fingerprint for watermarked content MUST include a ‘label’ (the unique fingerprint) and ‘text’.
• Object Contract: License plate
  This document describes the license plate of the Dukes of Hazzard’s General Lee. A license plate object MUST include ‘licensePlate’ property, and MAY include some other properties. If the license plate is detected attached to an identifiable vehicle, then ‘motorVehicle’ type should be used instead.
• Object Contract: Logo
  This document describes the logos found in a test document. A logo object MUST include ‘label’, ‘confidence’ and ‘boundingPoly’, and MAY include some other properties.
• Object Contract: Motor vehicle
  This document describes the Dukes of Hazzard’s General Lee with the optional license plate. A motor vehicle MUST include ‘motorVehicle’ property with at least one identifying property in it, and MAY include other properties.
• Object Contract: Motor vehicle with referenceId, confidence and tags
  A motor vehicle object using the v2.1 additions: top-level referenceId and confidence, and a tags array for arbitrary key/value attributes with optional score.
• Object Contract: Named Entity
  Named entity objects refer to specific entities (individuals, locations, times, organizations, etc) in a document along with an identifying category for that entity. Named entities MUST include a ‘label’ for the entity and an ‘objectCategory’ for the identification, and MAY include some other properties.
• Object Contract: Object
  Object detection is the process of identifying objects within an image or video. Objects are the most general type of visual entity identification, and should be used only when more specific types (like ‘face’ or ‘licensePlate’) are not applicable. Objects SHOULD have a ‘boundingPoly’ and a ‘confidence’
• Object Contract: OCR (Optical Character Recognition)
  OCR is the extraction of text from images or video frames. OCR objects are a specialization of ‘text’ objects that include bounding polygons for the text. OCR objects MUST include ‘text’ and ‘boundingPoly’, and MAY include some other properties.
• Object Contract: Sound
  Sound identification may include music, sound effects, other audio content like silence or static. Sound objects MUST include a ‘label’, SHOULD include ‘confidence’, and MAY include some other properties.
• Object Contract: Speaker
  Speaker identification processes audio files and identifies individual speakers. The speaker object MUST include a ‘label’ to identify the speaker. The object MAY include some other properties as well. Typically, the speakers are identified within the context of one audio file, and are labeled something like ‘Speaker 1’ and ‘Speaker 2’ or simply ‘A’ and ‘B’, but if an external reference is available, the speaker may be identified by name or entityId.
• Object Contract: Text (extraction)
  Text objects may refer to text extracted from a file, like in the case of OCR (optical character recognition).
• Object Contract: Text (generation)
  Text objects may store generated text like this output from an input document containing ‘Provide a three sentence summary of The Hobbit’
• Object Contract: Text (analysis)
  Text objects may refer to text accompanied by an analysis of deeper meaning, like sentiment, intent, or emotional tone.
• Object Contract: Text (transcription)
  Text objects may contain transcriptions of spoken recordings. Note that the ‘transcript’ Capability Contract requires word-by-word transcription, whereas a transcription stored in ‘text’ may be more generalized.
• Object Contract: Text (translation)
  Text objects may refer to text translated from another file, like in the case of translation of a Memo from German to English.