Partilhar via


Image Tagging (U-SQL)

Summary

Cognitive image tagging functions return information about visual content found in an image. Use tagging, descriptions, and domain-specific models to identify content and label content with confidence. There are two ways in U-SQL to extract contents and its confidence from an image:

  • ImageTagger
  • ImageTagsExtractor

Important

Use an Extractor rather than a Processor when you have images larger than 4 MB.

Arguments

ImageTagger | ImageTagsExtractor(
string numCol = "NumObjects", string tagCol = "Tags")

ImageTagger and ImageTagsExtractor will operate on a JPEG file and return a rowset with two columns called NumObjects of type int that provides the number of detected objects in the image and Tags that returns a SQL.MAP<string, float?> instance which contains the set of object tags as keys with a floating point value indicating the confidence of each of the tags as values.

Examples

A. ImageTagger

Extract tags from the image using Image tagging Processor

REFERENCE ASSEMBLY ImageCommon; 
REFERENCE ASSEMBLY ImageTagging;

@TaggedObjects =
    PROCESS dbo.myImages    
    PRODUCE FileName,
            NumObjects int,
            Tags SQL.MAP<string, float?>
    READONLY FileName
    USING new Cognition.Vision.ImageTagger();

Perform various queries against @TaggedObjects created above

// A record will not be returned for FileName if Tags is empty for that record
@result1A = 
    SELECT  FileName,
            NumObjects,
            T.key AS Content,
            T.value AS Confidence
    FROM    @TaggedObjects
    CROSS APPLY EXPLODE (Tags) AS T(key, value);


// Use the edited code below if you want a default value to be returned for a FileName that has no Tags
@result1B = 
    SELECT  FileName,
            NumObjects,
            T.key AS Content,
            T.value AS Confidence
    FROM    @TaggedObjects
    CROSS APPLY EXPLODE ((Equals(Tags.Count, 0)) ? new SQL.MAP<string, float?>{{"None", 0.0f}} : Tags) AS T(key, value);

// Alternative output
@result2 =
    SELECT FileName,
           NumObjects,
           String.Join("; ", Tags.Select(x => String.Format("{0}:{1}", x.Key, x.Value))) AS TagsString
    FROM @TaggedObjects;

// Filtering
@result3 =
    SELECT FileName,
           NumObjects,
           T.key AS Content,
           T.value AS Confidence
    FROM @TaggedObjects
         CROSS APPLY
             EXPLODE(Tags) AS T(key, value)
    WHERE T.key == "athletic game" 
    AND T.value >= 0.85f;


OUTPUT @result1A
TO "/ReferenceGuide/Cognition/Vision/ImageTagger1A.txt"
ORDER BY FileName
USING Outputters.Tsv(outputHeader: true);

OUTPUT @result1B
TO "/ReferenceGuide/Cognition/Vision/ImageTagger1B.txt"
ORDER BY FileName
USING Outputters.Tsv(outputHeader: true);

OUTPUT @result2
TO "/ReferenceGuide/Cognition/Vision/ImageTagger2.txt"
ORDER BY FileName
USING Outputters.Tsv();

OUTPUT @result3
TO "/ReferenceGuide/Cognition/Vision/ImageTagger3.txt"
ORDER BY FileName
USING Outputters.Tsv(outputHeader: true);

B. ImageTagsExtractor

Extract the tags from each image using image tagging Extractor

REFERENCE ASSEMBLY ImageCommon;   
REFERENCE ASSEMBLY ImageTagging;

@tags =
    EXTRACT FileName string, 
            NumObjects int,
            Tags SQL.MAP<string, float?>
    FROM @"/Samples/Images/{FileName}.jpg"
    USING new Cognition.Vision.ImageTagsExtractor();

Perform various queries against @tags created above

// Merge the Map into a string for outputting
@tags_serialized =
    SELECT FileName,
           NumObjects,
           String.Join(";", Tags.Select(x => String.Format("{0}:{1}", x.Key, x.Value))) AS TagsString
    FROM @tags;

// Flatten results
@result2 = 
    SELECT  FileName,
            NumObjects,
            T.key AS Content,
            T.value AS Confidence
    FROM    @tags
    CROSS APPLY EXPLODE (Tags) AS T(key, value);

// alternative view
@result3 =
    SELECT  FileName,
            NumObjects,
            string.Join(",", Tags) AS Tags
    FROM @tags;

OUTPUT @tags_serialized
TO "/ReferenceGuide/Cognition/Vision/ImageTagsExtractor.txt"
USING Outputters.Tsv();

OUTPUT @result2
TO "/ReferenceGuide/Cognition/Vision/ImageTagsExtractor2.txt"
USING Outputters.Tsv(outputHeader: true);

OUTPUT @result3
TO "/ReferenceGuide/Cognition/Vision/ImageTagsExtractor3.txt"
USING Outputters.Tsv();

See Also