# Generated by the protocol buffer compiler. DO NOT EDIT! # source: google/cloud/videointelligence/v1beta2/video_intelligence.proto require 'google/protobuf' require 'google/api/annotations_pb' require 'google/api/client_pb' require 'google/api/field_behavior_pb' require 'google/longrunning/operations_pb' require 'google/protobuf/duration_pb' require 'google/protobuf/timestamp_pb' require 'google/rpc/status_pb' Google::Protobuf::DescriptorPool.generated_pool.build do add_message "google.cloud.videointelligence.v1beta2.AnnotateVideoRequest" do optional :input_uri, :string, 1 optional :input_content, :bytes, 6 repeated :features, :enum, 2, "google.cloud.videointelligence.v1beta2.Feature" optional :video_context, :message, 3, "google.cloud.videointelligence.v1beta2.VideoContext" optional :output_uri, :string, 4 optional :location_id, :string, 5 end add_message "google.cloud.videointelligence.v1beta2.VideoContext" do repeated :segments, :message, 1, "google.cloud.videointelligence.v1beta2.VideoSegment" optional :label_detection_config, :message, 2, "google.cloud.videointelligence.v1beta2.LabelDetectionConfig" optional :shot_change_detection_config, :message, 3, "google.cloud.videointelligence.v1beta2.ShotChangeDetectionConfig" optional :explicit_content_detection_config, :message, 4, "google.cloud.videointelligence.v1beta2.ExplicitContentDetectionConfig" optional :face_detection_config, :message, 5, "google.cloud.videointelligence.v1beta2.FaceDetectionConfig" end add_message "google.cloud.videointelligence.v1beta2.LabelDetectionConfig" do optional :label_detection_mode, :enum, 1, "google.cloud.videointelligence.v1beta2.LabelDetectionMode" optional :stationary_camera, :bool, 2 optional :model, :string, 3 end add_message "google.cloud.videointelligence.v1beta2.ShotChangeDetectionConfig" do optional :model, :string, 1 end add_message "google.cloud.videointelligence.v1beta2.ExplicitContentDetectionConfig" do optional :model, :string, 1 end add_message "google.cloud.videointelligence.v1beta2.FaceDetectionConfig" do optional :model, :string, 1 optional :include_bounding_boxes, :bool, 2 end add_message "google.cloud.videointelligence.v1beta2.VideoSegment" do optional :start_time_offset, :message, 1, "google.protobuf.Duration" optional :end_time_offset, :message, 2, "google.protobuf.Duration" end add_message "google.cloud.videointelligence.v1beta2.LabelSegment" do optional :segment, :message, 1, "google.cloud.videointelligence.v1beta2.VideoSegment" optional :confidence, :float, 2 end add_message "google.cloud.videointelligence.v1beta2.LabelFrame" do optional :time_offset, :message, 1, "google.protobuf.Duration" optional :confidence, :float, 2 end add_message "google.cloud.videointelligence.v1beta2.Entity" do optional :entity_id, :string, 1 optional :description, :string, 2 optional :language_code, :string, 3 end add_message "google.cloud.videointelligence.v1beta2.LabelAnnotation" do optional :entity, :message, 1, "google.cloud.videointelligence.v1beta2.Entity" repeated :category_entities, :message, 2, "google.cloud.videointelligence.v1beta2.Entity" repeated :segments, :message, 3, "google.cloud.videointelligence.v1beta2.LabelSegment" repeated :frames, :message, 4, "google.cloud.videointelligence.v1beta2.LabelFrame" end add_message "google.cloud.videointelligence.v1beta2.ExplicitContentFrame" do optional :time_offset, :message, 1, "google.protobuf.Duration" optional :pornography_likelihood, :enum, 2, "google.cloud.videointelligence.v1beta2.Likelihood" end add_message "google.cloud.videointelligence.v1beta2.ExplicitContentAnnotation" do repeated :frames, :message, 1, "google.cloud.videointelligence.v1beta2.ExplicitContentFrame" end add_message "google.cloud.videointelligence.v1beta2.NormalizedBoundingBox" do optional :left, :float, 1 optional :top, :float, 2 optional :right, :float, 3 optional :bottom, :float, 4 end add_message "google.cloud.videointelligence.v1beta2.FaceSegment" do optional :segment, :message, 1, "google.cloud.videointelligence.v1beta2.VideoSegment" end add_message "google.cloud.videointelligence.v1beta2.FaceFrame" do repeated :normalized_bounding_boxes, :message, 1, "google.cloud.videointelligence.v1beta2.NormalizedBoundingBox" optional :time_offset, :message, 2, "google.protobuf.Duration" end add_message "google.cloud.videointelligence.v1beta2.FaceAnnotation" do optional :thumbnail, :bytes, 1 repeated :segments, :message, 2, "google.cloud.videointelligence.v1beta2.FaceSegment" repeated :frames, :message, 3, "google.cloud.videointelligence.v1beta2.FaceFrame" end add_message "google.cloud.videointelligence.v1beta2.VideoAnnotationResults" do optional :input_uri, :string, 1 repeated :segment_label_annotations, :message, 2, "google.cloud.videointelligence.v1beta2.LabelAnnotation" repeated :shot_label_annotations, :message, 3, "google.cloud.videointelligence.v1beta2.LabelAnnotation" repeated :frame_label_annotations, :message, 4, "google.cloud.videointelligence.v1beta2.LabelAnnotation" repeated :face_annotations, :message, 5, "google.cloud.videointelligence.v1beta2.FaceAnnotation" repeated :shot_annotations, :message, 6, "google.cloud.videointelligence.v1beta2.VideoSegment" optional :explicit_annotation, :message, 7, "google.cloud.videointelligence.v1beta2.ExplicitContentAnnotation" optional :error, :message, 9, "google.rpc.Status" end add_message "google.cloud.videointelligence.v1beta2.AnnotateVideoResponse" do repeated :annotation_results, :message, 1, "google.cloud.videointelligence.v1beta2.VideoAnnotationResults" end add_message "google.cloud.videointelligence.v1beta2.VideoAnnotationProgress" do optional :input_uri, :string, 1 optional :progress_percent, :int32, 2 optional :start_time, :message, 3, "google.protobuf.Timestamp" optional :update_time, :message, 4, "google.protobuf.Timestamp" end add_message "google.cloud.videointelligence.v1beta2.AnnotateVideoProgress" do repeated :annotation_progress, :message, 1, "google.cloud.videointelligence.v1beta2.VideoAnnotationProgress" end add_enum "google.cloud.videointelligence.v1beta2.Feature" do value :FEATURE_UNSPECIFIED, 0 value :LABEL_DETECTION, 1 value :SHOT_CHANGE_DETECTION, 2 value :EXPLICIT_CONTENT_DETECTION, 3 value :FACE_DETECTION, 4 end add_enum "google.cloud.videointelligence.v1beta2.LabelDetectionMode" do value :LABEL_DETECTION_MODE_UNSPECIFIED, 0 value :SHOT_MODE, 1 value :FRAME_MODE, 2 value :SHOT_AND_FRAME_MODE, 3 end add_enum "google.cloud.videointelligence.v1beta2.Likelihood" do value :LIKELIHOOD_UNSPECIFIED, 0 value :VERY_UNLIKELY, 1 value :UNLIKELY, 2 value :POSSIBLE, 3 value :LIKELY, 4 value :VERY_LIKELY, 5 end end module Google module Cloud module VideoIntelligence end Videointelligence = VideoIntelligence unless const_defined? :Videointelligence end end module Google::Cloud::VideoIntelligence::V1beta2 AnnotateVideoRequest = Google::Protobuf::DescriptorPool.generated_pool.lookup("google.cloud.videointelligence.v1beta2.AnnotateVideoRequest").msgclass VideoContext = Google::Protobuf::DescriptorPool.generated_pool.lookup("google.cloud.videointelligence.v1beta2.VideoContext").msgclass LabelDetectionConfig = Google::Protobuf::DescriptorPool.generated_pool.lookup("google.cloud.videointelligence.v1beta2.LabelDetectionConfig").msgclass ShotChangeDetectionConfig = Google::Protobuf::DescriptorPool.generated_pool.lookup("google.cloud.videointelligence.v1beta2.ShotChangeDetectionConfig").msgclass ExplicitContentDetectionConfig = Google::Protobuf::DescriptorPool.generated_pool.lookup("google.cloud.videointelligence.v1beta2.ExplicitContentDetectionConfig").msgclass FaceDetectionConfig = Google::Protobuf::DescriptorPool.generated_pool.lookup("google.cloud.videointelligence.v1beta2.FaceDetectionConfig").msgclass VideoSegment = Google::Protobuf::DescriptorPool.generated_pool.lookup("google.cloud.videointelligence.v1beta2.VideoSegment").msgclass LabelSegment = Google::Protobuf::DescriptorPool.generated_pool.lookup("google.cloud.videointelligence.v1beta2.LabelSegment").msgclass LabelFrame = Google::Protobuf::DescriptorPool.generated_pool.lookup("google.cloud.videointelligence.v1beta2.LabelFrame").msgclass Entity = Google::Protobuf::DescriptorPool.generated_pool.lookup("google.cloud.videointelligence.v1beta2.Entity").msgclass LabelAnnotation = Google::Protobuf::DescriptorPool.generated_pool.lookup("google.cloud.videointelligence.v1beta2.LabelAnnotation").msgclass ExplicitContentFrame = Google::Protobuf::DescriptorPool.generated_pool.lookup("google.cloud.videointelligence.v1beta2.ExplicitContentFrame").msgclass ExplicitContentAnnotation = Google::Protobuf::DescriptorPool.generated_pool.lookup("google.cloud.videointelligence.v1beta2.ExplicitContentAnnotation").msgclass NormalizedBoundingBox = Google::Protobuf::DescriptorPool.generated_pool.lookup("google.cloud.videointelligence.v1beta2.NormalizedBoundingBox").msgclass FaceSegment = Google::Protobuf::DescriptorPool.generated_pool.lookup("google.cloud.videointelligence.v1beta2.FaceSegment").msgclass FaceFrame = Google::Protobuf::DescriptorPool.generated_pool.lookup("google.cloud.videointelligence.v1beta2.FaceFrame").msgclass FaceAnnotation = Google::Protobuf::DescriptorPool.generated_pool.lookup("google.cloud.videointelligence.v1beta2.FaceAnnotation").msgclass VideoAnnotationResults = Google::Protobuf::DescriptorPool.generated_pool.lookup("google.cloud.videointelligence.v1beta2.VideoAnnotationResults").msgclass AnnotateVideoResponse = Google::Protobuf::DescriptorPool.generated_pool.lookup("google.cloud.videointelligence.v1beta2.AnnotateVideoResponse").msgclass VideoAnnotationProgress = Google::Protobuf::DescriptorPool.generated_pool.lookup("google.cloud.videointelligence.v1beta2.VideoAnnotationProgress").msgclass AnnotateVideoProgress = Google::Protobuf::DescriptorPool.generated_pool.lookup("google.cloud.videointelligence.v1beta2.AnnotateVideoProgress").msgclass Feature = Google::Protobuf::DescriptorPool.generated_pool.lookup("google.cloud.videointelligence.v1beta2.Feature").enummodule LabelDetectionMode = Google::Protobuf::DescriptorPool.generated_pool.lookup("google.cloud.videointelligence.v1beta2.LabelDetectionMode").enummodule Likelihood = Google::Protobuf::DescriptorPool.generated_pool.lookup("google.cloud.videointelligence.v1beta2.Likelihood").enummodule end