package google.cloud.aiplatform.v1beta1.schema.predict.prediction

Get desktop application:
View/edit binary Protocol Buffers messages

Prediction output format for Image and Text Classification.

• repeated int64 ids = 1

  The resource IDs of the AnnotationSpecs that had been identified.

• repeated string display_names = 2

  The display names of the AnnotationSpecs that had been identified, order matches the IDs.

• repeated float confidences = 3

  The Model's confidences in correctness of the predicted IDs, higher value means higher confidence. Order matches the Ids.

Prediction output format for Image Object Detection.

• repeated int64 ids = 1

  The resource IDs of the AnnotationSpecs that had been identified, ordered by the confidence score descendingly.

• repeated string display_names = 2

  The display names of the AnnotationSpecs that had been identified, order matches the IDs.

• repeated float confidences = 3

  The Model's confidences in correctness of the predicted IDs, higher value means higher confidence. Order matches the Ids.

• repeated protobuf.ListValue bboxes = 4

  Bounding boxes, i.e. the rectangles over the image, that pinpoint the found AnnotationSpecs. Given in order that matches the IDs. Each bounding box is an array of 4 numbers `xMin`, `xMax`, `yMin`, and `yMax`, which represent the extremal coordinates of the box. They are relative to the image size, and the point 0,0 is in the top left of the image.

Prediction output format for Image Segmentation.

• string category_mask = 1

  A PNG image where each pixel in the mask represents the category in which the pixel in the original image was predicted to belong to. The size of this image will be the same as the original image. The mapping between the AnntoationSpec and the color can be found in model's metadata. The model will choose the most likely category and if none of the categories reach the confidence threshold, the pixel will be marked as background.

• string confidence_mask = 2

  A one channel image which is encoded as an 8bit lossless PNG. The size of the image will be the same as the original image. For a specific pixel, darker color means less confidence in correctness of the cateogry in the categoryMask for the corresponding pixel. Black means no confidence and white means complete confidence.

Prediction output format for Tabular Classification.

• repeated string classes = 1

  The name of the classes being classified, contains all possible values of the target column.

• repeated float scores = 2

  The model's confidence in each class being correct, higher value means higher confidence. The N-th score corresponds to the N-th class in classes.

Prediction output format for Tabular Regression.

• float value = 1

  The regression value.

• float lower_bound = 2

  The lower bound of the prediction interval.

• float upper_bound = 3

  The upper bound of the prediction interval.

Prediction output format for Text Extraction.

• repeated int64 ids = 1

  The resource IDs of the AnnotationSpecs that had been identified, ordered by the confidence score descendingly.

• repeated string display_names = 2

  The display names of the AnnotationSpecs that had been identified, order matches the IDs.

• repeated int64 text_segment_start_offsets = 3

  The start offsets, inclusive, of the text segment in which the AnnotationSpec has been identified. Expressed as a zero-based number of characters as measured from the start of the text snippet.

• repeated int64 text_segment_end_offsets = 4

  The end offsets, inclusive, of the text segment in which the AnnotationSpec has been identified. Expressed as a zero-based number of characters as measured from the start of the text snippet.

• repeated float confidences = 5

  The Model's confidences in correctness of the predicted IDs, higher value means higher confidence. Order matches the Ids.

Prediction output format for Text Sentiment

• int32 sentiment = 1

  The integer sentiment labels between 0 (inclusive) and sentimentMax label (inclusive), while 0 maps to the least positive sentiment and sentimentMax maps to the most positive one. The higher the score is, the more positive the sentiment in the text snippet is. Note: sentimentMax is an integer value between 1 (inclusive) and 10 (inclusive).

Prediction output format for Time Series Forecasting.

• float value = 1

  The regression value.

Prediction output format for Video Action Recognition.

• string id = 1

  The resource ID of the AnnotationSpec that had been identified.

• string display_name = 2

  The display name of the AnnotationSpec that had been identified.

• optional protobuf.Duration time_segment_start = 4

  The beginning, inclusive, of the video's time segment in which the AnnotationSpec has been identified. Expressed as a number of seconds as measured from the start of the video, with fractions up to a microsecond precision, and with "s" appended at the end.

• optional protobuf.Duration time_segment_end = 5

  The end, exclusive, of the video's time segment in which the AnnotationSpec has been identified. Expressed as a number of seconds as measured from the start of the video, with fractions up to a microsecond precision, and with "s" appended at the end.

• optional protobuf.FloatValue confidence = 6

  The Model's confidence in correction of this prediction, higher value means higher confidence.

Prediction output format for Video Classification.

• string id = 1

  The resource ID of the AnnotationSpec that had been identified.

• string display_name = 2

  The display name of the AnnotationSpec that had been identified.

• string type = 3

  The type of the prediction. The requested types can be configured via parameters. This will be one of - segment-classification - shot-classification - one-sec-interval-classification

• optional protobuf.Duration time_segment_start = 4

  The beginning, inclusive, of the video's time segment in which the AnnotationSpec has been identified. Expressed as a number of seconds as measured from the start of the video, with fractions up to a microsecond precision, and with "s" appended at the end. Note that for 'segment-classification' prediction type, this equals the original 'timeSegmentStart' from the input instance, for other types it is the start of a shot or a 1 second interval respectively.

• optional protobuf.Duration time_segment_end = 5

  The end, exclusive, of the video's time segment in which the AnnotationSpec has been identified. Expressed as a number of seconds as measured from the start of the video, with fractions up to a microsecond precision, and with "s" appended at the end. Note that for 'segment-classification' prediction type, this equals the original 'timeSegmentEnd' from the input instance, for other types it is the end of a shot or a 1 second interval respectively.

• optional protobuf.FloatValue confidence = 6

  The Model's confidence in correction of this prediction, higher value means higher confidence.

Prediction output format for Video Object Tracking.

• string id = 1

  The resource ID of the AnnotationSpec that had been identified.

• string display_name = 2

  The display name of the AnnotationSpec that had been identified.

• optional protobuf.Duration time_segment_start = 3

  The beginning, inclusive, of the video's time segment in which the object instance has been detected. Expressed as a number of seconds as measured from the start of the video, with fractions up to a microsecond precision, and with "s" appended at the end.

• optional protobuf.Duration time_segment_end = 4

  The end, inclusive, of the video's time segment in which the object instance has been detected. Expressed as a number of seconds as measured from the start of the video, with fractions up to a microsecond precision, and with "s" appended at the end.

• optional protobuf.FloatValue confidence = 5

  The Model's confidence in correction of this prediction, higher value means higher confidence.

• repeated VideoObjectTrackingPredictionResult.Frame frames = 6

  All of the frames of the video in which a single object instance has been detected. The bounding boxes in the frames identify the same object.

The fields `xMin`, `xMax`, `yMin`, and `yMax` refer to a bounding box, i.e. the rectangle over the video frame pinpointing the found AnnotationSpec. The coordinates are relative to the frame size, and the point 0,0 is in the top left of the frame.

Used in: VideoObjectTrackingPredictionResult

• optional protobuf.Duration time_offset = 1

  A time (frame) of a video in which the object has been detected. Expressed as a number of seconds as measured from the start of the video, with fractions up to a microsecond precision, and with "s" appended at the end.

• optional protobuf.FloatValue x_min = 2

  The leftmost coordinate of the bounding box.

• optional protobuf.FloatValue x_max = 3

  The rightmost coordinate of the bounding box.

• optional protobuf.FloatValue y_min = 4

  The topmost coordinate of the bounding box.

• optional protobuf.FloatValue y_max = 5

  The bottommost coordinate of the bounding box.