Release Notes

This page documents production updates to Video Intelligence API. We recommend that Video Intelligence API developers periodically check this list for any new announcements.

October 26, 2018

Object Tracking and Text Detection (OCR) Beta

  • Object tracking is available as a beta feature. Learn more.

  • Text detection (OCR) is available as a beta feature. Learn more.

August 8, 2018

Speech Transcription Beta

Speech transcription is available as a beta feature. Learn more.

June 27, 2018

Label Detection Model Update

An improved label detection model is now available. The new model:

  • Leverages audio content in videos to improve label detection.
  • Is trained using more features and better calibration ground truth.

To instruct the Cloud Video Intelligence API to use the new label detection model when servicing your annotation request, set the model field of your LabelDetectionConfig to "builtin/latest" as shown in the following example:

{
  "inputUri": "gs://cloud-ml-sandbox/video/chicago.mp4",
  "features": ["LABEL_DETECTION"],
  "videoContext": {
    "labelDetectionConfig": {
      "model": "builtin/latest"
    },
  }
}
  

We'll support both the current model and the new model the next 90 days. After 90 days the current label detection model will be deprecated and only the new label detection model will be used for all label detection requests.

November 30, 2017

Video Intelligence API v1 General Availability Release

Video Intelligence API is GA: The Video Intelligence API has graduated out of beta and has reached v1. All API endpoints are updated to use https://videointelligence.googleapis.com/v1/.

September 21, 2017

Video Intelligence API v1beta2 Release

Explicit content detection: SafeSearch has been renamed to Explicit Content Detection. Explicit Content Detection inspects an input video for frame-level imagery that could be considered adult content.

Explicit Content Detection is performed by using the annotate method and specifying an EXPLICIT_CONTENT_DETECTION request.

The ExplicitContentAnnotation and ExplicitContentFrame types replace the SafeSearchAnnotation type. The adult field has been renamed to pornographyLikelihood and the time field has been renamed to timeOffset. The spoof, medical, violent, and racy fields have also been removed. The timeOffset field returns a value of type Duration instead of type int64.

The labelAnnotations field returned in the response for a video annotation request has been replaced with the segmentLabelAnnotations, shotLabelAnnotations, and frameLabelAnnotations fields. This provides specific label annotations for each level of the video. The LabelLevel enum has been removed.

All annotations have been updated to return an array of LabelFrame types, an array of LabelSegment types, a list of entities, and a list of entity categories. Each LabelFrame and LabelSegment includes a confidence value.

The field description has been replaced with an Entity type, which includes both the description and an entity_id field. You can use the entity id to find more information for some entities in the Google Knowledge Graph Search API.

The LabelDetectionMode enum remains unchanged and can be set as the label_detection_mode field of the LabelDetectionConfig for your request, along with the stationary_camera field. This configurations applies only to Labels. Other video intelligence features do not yet have yet any feature-specific configuration options.

The VideoSegment type is now used only in the context configuration (VideoContext) of the AnnotateVideoRequest type to allow you to pass multiple video segments in a request.

June 26, 2017

SafeSearch detection

SafeSearch detection: SafeSearch inspects an input video for frame-level imagery that could be considered adult content.

SafeSearch is performed through the annotate method using a SAFE_SEARCH_DETECTION request.

May 18, 2017

Cloud Video Intelligence API Beta Release

Shot change detection: Shot detection inspects an input video, and detects changes in shots (scenes) that occur throughout the length of the video or video segment. Shot detection is performed through the annotate method using a SHOT_CHANGE_DETECTION request.

Label detection: Label detection inspects an input video and detects entities that occur throughout the length of the video or video segment. Label detection is performed through the annotate method using a LABEL_DETECTION request.

Var denne side nyttig? Giv os en anmeldelse af den:

Send feedback om...

Cloud Video Intelligence API