kpt-grpc-demo/google/cloud/videointelligence/v1/video_intelligence.proto

// Copyright 2020 Google LLC
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
//     http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.

syntax = "proto3";

package google.cloud.videointelligence.v1;

import "google/api/annotations.proto";
import "google/api/client.proto";
import "google/api/field_behavior.proto";
import "google/longrunning/operations.proto";
import "google/protobuf/duration.proto";
import "google/protobuf/timestamp.proto";
import "google/rpc/status.proto";

option csharp_namespace = "Google.Cloud.VideoIntelligence.V1";
option go_package = "google.golang.org/genproto/googleapis/cloud/videointelligence/v1;videointelligence";
option java_multiple_files = true;
option java_outer_classname = "VideoIntelligenceServiceProto";
option java_package = "com.google.cloud.videointelligence.v1";
option php_namespace = "Google\\Cloud\\VideoIntelligence\\V1";
option ruby_package = "Google::Cloud::VideoIntelligence::V1";

// Service that implements the Video Intelligence API.
service VideoIntelligenceService {
  option (google.api.default_host) = "videointelligence.googleapis.com";
  option (google.api.oauth_scopes) =
      "https://www.googleapis.com/auth/cloud-platform";

  // Performs asynchronous video annotation. Progress and results can be
  // retrieved through the `google.longrunning.Operations` interface.
  // `Operation.metadata` contains `AnnotateVideoProgress` (progress).
  // `Operation.response` contains `AnnotateVideoResponse` (results).
  rpc AnnotateVideo(AnnotateVideoRequest)
      returns (google.longrunning.Operation) {
    option (google.api.http) = {
      post: "/v1/videos:annotate"
      body: "*"
    };
    option (google.api.method_signature) = "input_uri,features";
    option (google.longrunning.operation_info) = {
      response_type: "AnnotateVideoResponse"
      metadata_type: "AnnotateVideoProgress"
    };
  }
}

// Video annotation request.
message AnnotateVideoRequest {
  // Input video location. Currently, only
  // [Cloud Storage](https://cloud.google.com/storage/) URIs are
  // supported. URIs must be specified in the following format:
  // `gs://bucket-id/object-id` (other URI formats return
  // [google.rpc.Code.INVALID_ARGUMENT][google.rpc.Code.INVALID_ARGUMENT]). For
  // more information, see [Request
  // URIs](https://cloud.google.com/storage/docs/request-endpoints). To identify
  // multiple videos, a video URI may include wildcards in the `object-id`.
  // Supported wildcards: '*' to match 0 or more characters;
  // '?' to match 1 character. If unset, the input video should be embedded
  // in the request as `input_content`. If set, `input_content` must be unset.
  string input_uri = 1;

  // The video data bytes.
  // If unset, the input video(s) should be specified via the `input_uri`.
  // If set, `input_uri` must be unset.
  bytes input_content = 6;

  // Required. Requested video annotation features.
  repeated Feature features = 2 [(google.api.field_behavior) = REQUIRED];

  // Additional video context and/or feature-specific parameters.
  VideoContext video_context = 3;

  // Optional. Location where the output (in JSON format) should be stored.
  // Currently, only [Cloud Storage](https://cloud.google.com/storage/)
  // URIs are supported. These must be specified in the following format:
  // `gs://bucket-id/object-id` (other URI formats return
  // [google.rpc.Code.INVALID_ARGUMENT][google.rpc.Code.INVALID_ARGUMENT]). For
  // more information, see [Request
  // URIs](https://cloud.google.com/storage/docs/request-endpoints).
  string output_uri = 4 [(google.api.field_behavior) = OPTIONAL];

  // Optional. Cloud region where annotation should take place. Supported cloud
  // regions are: `us-east1`, `us-west1`, `europe-west1`, `asia-east1`. If no
  // region is specified, the region will be determined based on video file
  // location.
  string location_id = 5 [(google.api.field_behavior) = OPTIONAL];
}

// Video context and/or feature-specific parameters.
message VideoContext {
  // Video segments to annotate. The segments may overlap and are not required
  // to be contiguous or span the whole video. If unspecified, each video is
  // treated as a single segment.
  repeated VideoSegment segments = 1;

  // Config for LABEL_DETECTION.
  LabelDetectionConfig label_detection_config = 2;

  // Config for SHOT_CHANGE_DETECTION.
  ShotChangeDetectionConfig shot_change_detection_config = 3;

  // Config for EXPLICIT_CONTENT_DETECTION.
  ExplicitContentDetectionConfig explicit_content_detection_config = 4;

  // Config for FACE_DETECTION.
  FaceDetectionConfig face_detection_config = 5;

  // Config for SPEECH_TRANSCRIPTION.
  SpeechTranscriptionConfig speech_transcription_config = 6;

  // Config for TEXT_DETECTION.
  TextDetectionConfig text_detection_config = 8;

  // Config for PERSON_DETECTION.
  PersonDetectionConfig person_detection_config = 11;

  // Config for OBJECT_TRACKING.
  ObjectTrackingConfig object_tracking_config = 13;
}

// Video annotation feature.
enum Feature {
  // Unspecified.
  FEATURE_UNSPECIFIED = 0;

  // Label detection. Detect objects, such as dog or flower.
  LABEL_DETECTION = 1;

  // Shot change detection.
  SHOT_CHANGE_DETECTION = 2;

  // Explicit content detection.
  EXPLICIT_CONTENT_DETECTION = 3;

  // Human face detection.
  FACE_DETECTION = 4;

  // Speech transcription.
  SPEECH_TRANSCRIPTION = 6;

  // OCR text detection and tracking.
  TEXT_DETECTION = 7;

  // Object detection and tracking.
  OBJECT_TRACKING = 9;

  // Logo detection, tracking, and recognition.
  LOGO_RECOGNITION = 12;

  // Person detection.
  PERSON_DETECTION = 14;
}

// Label detection mode.
enum LabelDetectionMode {
  // Unspecified.
  LABEL_DETECTION_MODE_UNSPECIFIED = 0;

  // Detect shot-level labels.
  SHOT_MODE = 1;

  // Detect frame-level labels.
  FRAME_MODE = 2;

  // Detect both shot-level and frame-level labels.
  SHOT_AND_FRAME_MODE = 3;
}

// Bucketized representation of likelihood.
enum Likelihood {
  // Unspecified likelihood.
  LIKELIHOOD_UNSPECIFIED = 0;

  // Very unlikely.
  VERY_UNLIKELY = 1;

  // Unlikely.
  UNLIKELY = 2;

  // Possible.
  POSSIBLE = 3;

  // Likely.
  LIKELY = 4;

  // Very likely.
  VERY_LIKELY = 5;
}

// Config for LABEL_DETECTION.
message LabelDetectionConfig {
  // What labels should be detected with LABEL_DETECTION, in addition to
  // video-level labels or segment-level labels.
  // If unspecified, defaults to `SHOT_MODE`.
  LabelDetectionMode label_detection_mode = 1;

  // Whether the video has been shot from a stationary (i.e., non-moving)
  // camera. When set to true, might improve detection accuracy for moving
  // objects. Should be used with `SHOT_AND_FRAME_MODE` enabled.
  bool stationary_camera = 2;

  // Model to use for label detection.
  // Supported values: "builtin/stable" (the default if unset) and
  // "builtin/latest".
  string model = 3;

  // The confidence threshold we perform filtering on the labels from
  // frame-level detection. If not set, it is set to 0.4 by default. The valid
  // range for this threshold is [0.1, 0.9]. Any value set outside of this
  // range will be clipped.
  // Note: For best results, follow the default threshold. We will update
  // the default threshold everytime when we release a new model.
  float frame_confidence_threshold = 4;

  // The confidence threshold we perform filtering on the labels from
  // video-level and shot-level detections. If not set, it's set to 0.3 by
  // default. The valid range for this threshold is [0.1, 0.9]. Any value set
  // outside of this range will be clipped.
  // Note: For best results, follow the default threshold. We will update
  // the default threshold everytime when we release a new model.
  float video_confidence_threshold = 5;
}

// Config for SHOT_CHANGE_DETECTION.
message ShotChangeDetectionConfig {
  // Model to use for shot change detection.
  // Supported values: "builtin/stable" (the default if unset) and
  // "builtin/latest".
  string model = 1;
}

// Config for OBJECT_TRACKING.
message ObjectTrackingConfig {
  // Model to use for object tracking.
  // Supported values: "builtin/stable" (the default if unset) and
  // "builtin/latest".
  string model = 1;
}

// Config for FACE_DETECTION.
message FaceDetectionConfig {
  // Model to use for face detection.
  // Supported values: "builtin/stable" (the default if unset) and
  // "builtin/latest".
  string model = 1;

  // Whether bounding boxes are included in the face annotation output.
  bool include_bounding_boxes = 2;

  // Whether to enable face attributes detection, such as glasses, dark_glasses,
  // mouth_open etc. Ignored if 'include_bounding_boxes' is set to false.
  bool include_attributes = 5;
}

// Config for PERSON_DETECTION.
message PersonDetectionConfig {
  // Whether bounding boxes are included in the person detection annotation
  // output.
  bool include_bounding_boxes = 1;

  // Whether to enable pose landmarks detection. Ignored if
  // 'include_bounding_boxes' is set to false.
  bool include_pose_landmarks = 2;

  // Whether to enable person attributes detection, such as cloth color (black,
  // blue, etc), type (coat, dress, etc), pattern (plain, floral, etc), hair,
  // etc.
  // Ignored if 'include_bounding_boxes' is set to false.
  bool include_attributes = 3;
}

// Config for EXPLICIT_CONTENT_DETECTION.
message ExplicitContentDetectionConfig {
  // Model to use for explicit content detection.
  // Supported values: "builtin/stable" (the default if unset) and
  // "builtin/latest".
  string model = 1;
}

// Config for TEXT_DETECTION.
message TextDetectionConfig {
  // Language hint can be specified if the language to be detected is known a
  // priori. It can increase the accuracy of the detection. Language hint must
  // be language code in BCP-47 format.
  //
  // Automatic language detection is performed if no hint is provided.
  repeated string language_hints = 1;

  // Model to use for text detection.
  // Supported values: "builtin/stable" (the default if unset) and
  // "builtin/latest".
  string model = 2;
}

// Video segment.
message VideoSegment {
  // Time-offset, relative to the beginning of the video,
  // corresponding to the start of the segment (inclusive).
  google.protobuf.Duration start_time_offset = 1;

  // Time-offset, relative to the beginning of the video,
  // corresponding to the end of the segment (inclusive).
  google.protobuf.Duration end_time_offset = 2;
}

// Video segment level annotation results for label detection.
message LabelSegment {
  // Video segment where a label was detected.
  VideoSegment segment = 1;

  // Confidence that the label is accurate. Range: [0, 1].
  float confidence = 2;
}

// Video frame level annotation results for label detection.
message LabelFrame {
  // Time-offset, relative to the beginning of the video, corresponding to the
  // video frame for this location.
  google.protobuf.Duration time_offset = 1;

  // Confidence that the label is accurate. Range: [0, 1].
  float confidence = 2;
}

// Detected entity from video analysis.
message Entity {
  // Opaque entity ID. Some IDs may be available in
  // [Google Knowledge Graph Search
  // API](https://developers.google.com/knowledge-graph/).
  string entity_id = 1;

  // Textual description, e.g., `Fixed-gear bicycle`.
  string description = 2;

  // Language code for `description` in BCP-47 format.
  string language_code = 3;
}

// Label annotation.
message LabelAnnotation {
  // Detected entity.
  Entity entity = 1;

  // Common categories for the detected entity.
  // For example, when the label is `Terrier`, the category is likely `dog`. And
  // in some cases there might be more than one categories e.g., `Terrier` could
  // also be a `pet`.
  repeated Entity category_entities = 2;

  // All video segments where a label was detected.
  repeated LabelSegment segments = 3;

  // All video frames where a label was detected.
  repeated LabelFrame frames = 4;

  // Feature version.
  string version = 5;
}

// Video frame level annotation results for explicit content.
message ExplicitContentFrame {
  // Time-offset, relative to the beginning of the video, corresponding to the
  // video frame for this location.
  google.protobuf.Duration time_offset = 1;

  // Likelihood of the pornography content..
  Likelihood pornography_likelihood = 2;
}

// Explicit content annotation (based on per-frame visual signals only).
// If no explicit content has been detected in a frame, no annotations are
// present for that frame.
message ExplicitContentAnnotation {
  // All video frames where explicit content was detected.
  repeated ExplicitContentFrame frames = 1;

  // Feature version.
  string version = 2;
}

// Normalized bounding box.
// The normalized vertex coordinates are relative to the original image.
// Range: [0, 1].
message NormalizedBoundingBox {
  // Left X coordinate.
  float left = 1;

  // Top Y coordinate.
  float top = 2;

  // Right X coordinate.
  float right = 3;

  // Bottom Y coordinate.
  float bottom = 4;
}

// Face detection annotation.
message FaceDetectionAnnotation {
  // The face tracks with attributes.
  repeated Track tracks = 3;

  // The thumbnail of a person's face.
  bytes thumbnail = 4;

  // Feature version.
  string version = 5;
}

// Person detection annotation per video.
message PersonDetectionAnnotation {
  // The detected tracks of a person.
  repeated Track tracks = 1;

  // Feature version.
  string version = 2;
}

// Video segment level annotation results for face detection.
message FaceSegment {
  // Video segment where a face was detected.
  VideoSegment segment = 1;
}

// Deprecated. No effect.
message FaceFrame {
  option deprecated = true;

  // Normalized Bounding boxes in a frame.
  // There can be more than one boxes if the same face is detected in multiple
  // locations within the current frame.
  repeated NormalizedBoundingBox normalized_bounding_boxes = 1;

  // Time-offset, relative to the beginning of the video,
  // corresponding to the video frame for this location.
  google.protobuf.Duration time_offset = 2;
}

// Deprecated. No effect.
message FaceAnnotation {
  option deprecated = true;

  // Thumbnail of a representative face view (in JPEG format).
  bytes thumbnail = 1;

  // All video segments where a face was detected.
  repeated FaceSegment segments = 2;

  // All video frames where a face was detected.
  repeated FaceFrame frames = 3;
}

// For tracking related features.
// An object at time_offset with attributes, and located with
// normalized_bounding_box.
message TimestampedObject {
  // Normalized Bounding box in a frame, where the object is located.
  NormalizedBoundingBox normalized_bounding_box = 1;

  // Time-offset, relative to the beginning of the video,
  // corresponding to the video frame for this object.
  google.protobuf.Duration time_offset = 2;

  // Optional. The attributes of the object in the bounding box.
  repeated DetectedAttribute attributes = 3
      [(google.api.field_behavior) = OPTIONAL];

  // Optional. The detected landmarks.
  repeated DetectedLandmark landmarks = 4
      [(google.api.field_behavior) = OPTIONAL];
}

// A track of an object instance.
message Track {
  // Video segment of a track.
  VideoSegment segment = 1;

  // The object with timestamp and attributes per frame in the track.
  repeated TimestampedObject timestamped_objects = 2;

  // Optional. Attributes in the track level.
  repeated DetectedAttribute attributes = 3
      [(google.api.field_behavior) = OPTIONAL];

  // Optional. The confidence score of the tracked object.
  float confidence = 4 [(google.api.field_behavior) = OPTIONAL];
}

// A generic detected attribute represented by name in string format.
message DetectedAttribute {
  // The name of the attribute, for example, glasses, dark_glasses, mouth_open.
  // A full list of supported type names will be provided in the document.
  string name = 1;

  // Detected attribute confidence. Range [0, 1].
  float confidence = 2;

  // Text value of the detection result. For example, the value for "HairColor"
  // can be "black", "blonde", etc.
  string value = 3;
}

// A generic detected landmark represented by name in string format and a 2D
// location.
message DetectedLandmark {
  // The name of this landmark, for example, left_hand, right_shoulder.
  string name = 1;

  // The 2D point of the detected landmark using the normalized image
  // coordindate system. The normalized coordinates have the range from 0 to 1.
  NormalizedVertex point = 2;

  // The confidence score of the detected landmark. Range [0, 1].
  float confidence = 3;
}

// Annotation results for a single video.
message VideoAnnotationResults {
  // Video file location in
  // [Cloud Storage](https://cloud.google.com/storage/).
  string input_uri = 1;

  // Video segment on which the annotation is run.
  VideoSegment segment = 10;

  // Topical label annotations on video level or user-specified segment level.
  // There is exactly one element for each unique label.
  repeated LabelAnnotation segment_label_annotations = 2;

  // Presence label annotations on video level or user-specified segment level.
  // There is exactly one element for each unique label. Compared to the
  // existing topical `segment_label_annotations`, this field presents more
  // fine-grained, segment-level labels detected in video content and is made
  // available only when the client sets `LabelDetectionConfig.model` to
  // "builtin/latest" in the request.
  repeated LabelAnnotation segment_presence_label_annotations = 23;

  // Topical label annotations on shot level.
  // There is exactly one element for each unique label.
  repeated LabelAnnotation shot_label_annotations = 3;

  // Presence label annotations on shot level. There is exactly one element for
  // each unique label. Compared to the existing topical
  // `shot_label_annotations`, this field presents more fine-grained, shot-level
  // labels detected in video content and is made available only when the client
  // sets `LabelDetectionConfig.model` to "builtin/latest" in the request.
  repeated LabelAnnotation shot_presence_label_annotations = 24;

  // Label annotations on frame level.
  // There is exactly one element for each unique label.
  repeated LabelAnnotation frame_label_annotations = 4;

  // Deprecated. Please use `face_detection_annotations` instead.
  repeated FaceAnnotation face_annotations = 5 [deprecated = true];

  // Face detection annotations.
  repeated FaceDetectionAnnotation face_detection_annotations = 13;

  // Shot annotations. Each shot is represented as a video segment.
  repeated VideoSegment shot_annotations = 6;

  // Explicit content annotation.
  ExplicitContentAnnotation explicit_annotation = 7;

  // Speech transcription.
  repeated SpeechTranscription speech_transcriptions = 11;

  // OCR text detection and tracking.
  // Annotations for list of detected text snippets. Each will have list of
  // frame information associated with it.
  repeated TextAnnotation text_annotations = 12;

  // Annotations for list of objects detected and tracked in video.
  repeated ObjectTrackingAnnotation object_annotations = 14;

  // Annotations for list of logos detected, tracked and recognized in video.
  repeated LogoRecognitionAnnotation logo_recognition_annotations = 19;

  // Person detection annotations.
  repeated PersonDetectionAnnotation person_detection_annotations = 20;

  // If set, indicates an error. Note that for a single `AnnotateVideoRequest`
  // some videos may succeed and some may fail.
  google.rpc.Status error = 9;
}

// Video annotation response. Included in the `response`
// field of the `Operation` returned by the `GetOperation`
// call of the `google::longrunning::Operations` service.
message AnnotateVideoResponse {
  // Annotation results for all videos specified in `AnnotateVideoRequest`.
  repeated VideoAnnotationResults annotation_results = 1;
}

// Annotation progress for a single video.
message VideoAnnotationProgress {
  // Video file location in
  // [Cloud Storage](https://cloud.google.com/storage/).
  string input_uri = 1;

  // Approximate percentage processed thus far. Guaranteed to be
  // 100 when fully processed.
  int32 progress_percent = 2;

  // Time when the request was received.
  google.protobuf.Timestamp start_time = 3;

  // Time of the most recent update.
  google.protobuf.Timestamp update_time = 4;

  // Specifies which feature is being tracked if the request contains more than
  // one feature.
  Feature feature = 5;

  // Specifies which segment is being tracked if the request contains more than
  // one segment.
  VideoSegment segment = 6;
}

// Video annotation progress. Included in the `metadata`
// field of the `Operation` returned by the `GetOperation`
// call of the `google::longrunning::Operations` service.
message AnnotateVideoProgress {
  // Progress metadata for all videos specified in `AnnotateVideoRequest`.
  repeated VideoAnnotationProgress annotation_progress = 1;
}

// Config for SPEECH_TRANSCRIPTION.
message SpeechTranscriptionConfig {
  // Required. *Required* The language of the supplied audio as a
  // [BCP-47](https://www.rfc-editor.org/rfc/bcp/bcp47.txt) language tag.
  // Example: "en-US".
  // See [Language Support](https://cloud.google.com/speech/docs/languages)
  // for a list of the currently supported language codes.
  string language_code = 1 [(google.api.field_behavior) = REQUIRED];

  // Optional. Maximum number of recognition hypotheses to be returned.
  // Specifically, the maximum number of `SpeechRecognitionAlternative` messages
  // within each `SpeechTranscription`. The server may return fewer than
  // `max_alternatives`. Valid values are `0`-`30`. A value of `0` or `1` will
  // return a maximum of one. If omitted, will return a maximum of one.
  int32 max_alternatives = 2 [(google.api.field_behavior) = OPTIONAL];

  // Optional. If set to `true`, the server will attempt to filter out
  // profanities, replacing all but the initial character in each filtered word
  // with asterisks, e.g. "f***". If set to `false` or omitted, profanities
  // won't be filtered out.
  bool filter_profanity = 3 [(google.api.field_behavior) = OPTIONAL];

  // Optional. A means to provide context to assist the speech recognition.
  repeated SpeechContext speech_contexts = 4
      [(google.api.field_behavior) = OPTIONAL];

  // Optional. If 'true', adds punctuation to recognition result hypotheses.
  // This feature is only available in select languages. Setting this for
  // requests in other languages has no effect at all. The default 'false' value
  // does not add punctuation to result hypotheses. NOTE: "This is currently
  // offered as an experimental service, complimentary to all users. In the
  // future this may be exclusively available as a premium feature."
  bool enable_automatic_punctuation = 5
      [(google.api.field_behavior) = OPTIONAL];

  // Optional. For file formats, such as MXF or MKV, supporting multiple audio
  // tracks, specify up to two tracks. Default: track 0.
  repeated int32 audio_tracks = 6 [(google.api.field_behavior) = OPTIONAL];

  // Optional. If 'true', enables speaker detection for each recognized word in
  // the top alternative of the recognition result using a speaker_tag provided
  // in the WordInfo.
  // Note: When this is true, we send all the words from the beginning of the
  // audio for the top alternative in every consecutive response.
  // This is done in order to improve our speaker tags as our models learn to
  // identify the speakers in the conversation over time.
  bool enable_speaker_diarization = 7 [(google.api.field_behavior) = OPTIONAL];

  // Optional. If set, specifies the estimated number of speakers in the
  // conversation. If not set, defaults to '2'. Ignored unless
  // enable_speaker_diarization is set to true.
  int32 diarization_speaker_count = 8 [(google.api.field_behavior) = OPTIONAL];

  // Optional. If `true`, the top result includes a list of words and the
  // confidence for those words. If `false`, no word-level confidence
  // information is returned. The default is `false`.
  bool enable_word_confidence = 9 [(google.api.field_behavior) = OPTIONAL];
}

// Provides "hints" to the speech recognizer to favor specific words and phrases
// in the results.
message SpeechContext {
  // Optional. A list of strings containing words and phrases "hints" so that
  // the speech recognition is more likely to recognize them. This can be used
  // to improve the accuracy for specific words and phrases, for example, if
  // specific commands are typically spoken by the user. This can also be used
  // to add additional words to the vocabulary of the recognizer. See
  // [usage limits](https://cloud.google.com/speech/limits#content).
  repeated string phrases = 1 [(google.api.field_behavior) = OPTIONAL];
}

// A speech recognition result corresponding to a portion of the audio.
message SpeechTranscription {
  // May contain one or more recognition hypotheses (up to the maximum specified
  // in `max_alternatives`).  These alternatives are ordered in terms of
  // accuracy, with the top (first) alternative being the most probable, as
  // ranked by the recognizer.
  repeated SpeechRecognitionAlternative alternatives = 1;

  // Output only. The [BCP-47](https://www.rfc-editor.org/rfc/bcp/bcp47.txt)
  // language tag of the language in this result. This language code was
  // detected to have the most likelihood of being spoken in the audio.
  string language_code = 2 [(google.api.field_behavior) = OUTPUT_ONLY];
}

// Alternative hypotheses (a.k.a. n-best list).
message SpeechRecognitionAlternative {
  // Transcript text representing the words that the user spoke.
  string transcript = 1;

  // Output only. The confidence estimate between 0.0 and 1.0. A higher number
  // indicates an estimated greater likelihood that the recognized words are
  // correct. This field is set only for the top alternative.
  // This field is not guaranteed to be accurate and users should not rely on it
  // to be always provided.
  // The default of 0.0 is a sentinel value indicating `confidence` was not set.
  float confidence = 2 [(google.api.field_behavior) = OUTPUT_ONLY];

  // Output only. A list of word-specific information for each recognized word.
  // Note: When `enable_speaker_diarization` is set to true, you will see all
  // the words from the beginning of the audio.
  repeated WordInfo words = 3 [(google.api.field_behavior) = OUTPUT_ONLY];
}

// Word-specific information for recognized words. Word information is only
// included in the response when certain request parameters are set, such
// as `enable_word_time_offsets`.
message WordInfo {
  // Time offset relative to the beginning of the audio, and
  // corresponding to the start of the spoken word. This field is only set if
  // `enable_word_time_offsets=true` and only in the top hypothesis. This is an
  // experimental feature and the accuracy of the time offset can vary.
  google.protobuf.Duration start_time = 1;

  // Time offset relative to the beginning of the audio, and
  // corresponding to the end of the spoken word. This field is only set if
  // `enable_word_time_offsets=true` and only in the top hypothesis. This is an
  // experimental feature and the accuracy of the time offset can vary.
  google.protobuf.Duration end_time = 2;

  // The word corresponding to this set of information.
  string word = 3;

  // Output only. The confidence estimate between 0.0 and 1.0. A higher number
  // indicates an estimated greater likelihood that the recognized words are
  // correct. This field is set only for the top alternative.
  // This field is not guaranteed to be accurate and users should not rely on it
  // to be always provided.
  // The default of 0.0 is a sentinel value indicating `confidence` was not set.
  float confidence = 4 [(google.api.field_behavior) = OUTPUT_ONLY];

  // Output only. A distinct integer value is assigned for every speaker within
  // the audio. This field specifies which one of those speakers was detected to
  // have spoken this word. Value ranges from 1 up to diarization_speaker_count,
  // and is only set if speaker diarization is enabled.
  int32 speaker_tag = 5 [(google.api.field_behavior) = OUTPUT_ONLY];
}

// A vertex represents a 2D point in the image.
// NOTE: the normalized vertex coordinates are relative to the original image
// and range from 0 to 1.
message NormalizedVertex {
  // X coordinate.
  float x = 1;

  // Y coordinate.
  float y = 2;
}

// Normalized bounding polygon for text (that might not be aligned with axis).
// Contains list of the corner points in clockwise order starting from
// top-left corner. For example, for a rectangular bounding box:
// When the text is horizontal it might look like:
//         0----1
//         |    |
//         3----2
//
// When it's clockwise rotated 180 degrees around the top-left corner it
// becomes:
//         2----3
//         |    |
//         1----0
//
// and the vertex order will still be (0, 1, 2, 3). Note that values can be less
// than 0, or greater than 1 due to trignometric calculations for location of
// the box.
message NormalizedBoundingPoly {
  // Normalized vertices of the bounding polygon.
  repeated NormalizedVertex vertices = 1;
}

// Video segment level annotation results for text detection.
message TextSegment {
  // Video segment where a text snippet was detected.
  VideoSegment segment = 1;

  // Confidence for the track of detected text. It is calculated as the highest
  // over all frames where OCR detected text appears.
  float confidence = 2;

  // Information related to the frames where OCR detected text appears.
  repeated TextFrame frames = 3;
}

// Video frame level annotation results for text annotation (OCR).
// Contains information regarding timestamp and bounding box locations for the
// frames containing detected OCR text snippets.
message TextFrame {
  // Bounding polygon of the detected text for this frame.
  NormalizedBoundingPoly rotated_bounding_box = 1;

  // Timestamp of this frame.
  google.protobuf.Duration time_offset = 2;
}

// Annotations related to one detected OCR text snippet. This will contain the
// corresponding text, confidence value, and frame level information for each
// detection.
message TextAnnotation {
  // The detected text.
  string text = 1;

  // All video segments where OCR detected text appears.
  repeated TextSegment segments = 2;

  // Feature version.
  string version = 3;
}

// Video frame level annotations for object detection and tracking. This field
// stores per frame location, time offset, and confidence.
message ObjectTrackingFrame {
  // The normalized bounding box location of this object track for the frame.
  NormalizedBoundingBox normalized_bounding_box = 1;

  // The timestamp of the frame in microseconds.
  google.protobuf.Duration time_offset = 2;
}

// Annotations corresponding to one tracked object.
message ObjectTrackingAnnotation {
  // Different representation of tracking info in non-streaming batch
  // and streaming modes.
  oneof track_info {
    // Non-streaming batch mode ONLY.
    // Each object track corresponds to one video segment where it appears.
    VideoSegment segment = 3;

    // Streaming mode ONLY.
    // In streaming mode, we do not know the end time of a tracked object
    // before it is completed. Hence, there is no VideoSegment info returned.
    // Instead, we provide a unique identifiable integer track_id so that
    // the customers can correlate the results of the ongoing
    // ObjectTrackAnnotation of the same track_id over time.
    int64 track_id = 5;
  }

  // Entity to specify the object category that this track is labeled as.
  Entity entity = 1;

  // Object category's labeling confidence of this track.
  float confidence = 4;

  // Information corresponding to all frames where this object track appears.
  // Non-streaming batch mode: it may be one or multiple ObjectTrackingFrame
  // messages in frames.
  // Streaming mode: it can only be one ObjectTrackingFrame message in frames.
  repeated ObjectTrackingFrame frames = 2;

  // Feature version.
  string version = 6;
}

// Annotation corresponding to one detected, tracked and recognized logo class.
message LogoRecognitionAnnotation {
  // Entity category information to specify the logo class that all the logo
  // tracks within this LogoRecognitionAnnotation are recognized as.
  Entity entity = 1;

  // All logo tracks where the recognized logo appears. Each track corresponds
  // to one logo instance appearing in consecutive frames.
  repeated Track tracks = 2;

  // All video segments where the recognized logo appears. There might be
  // multiple instances of the same logo class appearing in one VideoSegment.
  repeated VideoSegment segments = 3;
}