Implement computer vision solutions 3.2 Design and implement multimodal understanding workflows
An agency uses a custom Content Understanding analyzer (baseAnalyzerId prebuilt-video) to…
Study guide skill: Implement video analysis workflows to process and interpret video segments
Question B7-008
An agency uses a custom Content Understanding analyzer (baseAnalyzerId prebuilt-video) to check short ads. Each ad needs one brand-safety verdict that looks across the whole ad and one full-length summary. A developer turned on custom segmentation, and the output now holds many segments, each with its own field values. What should you change?
Select one answer.
Answer: B
Content Understanding offers whole-video mode (enableSegment false) and custom segmentation (enableSegment true) on custom analyzers. Whole-video treats the entire file as one segment, which Microsoft lists for brand-safety checks anywhere in an ad and for full-length summaries. returnDetails only adds detail such as confidence scores and bounding boxes, and prebuilt-document is the base for document analyzers, not videos.
Deep dive
How it works
For video, a custom Content Understanding analyzer offers two segmentation modes through the enableSegment property:
- Whole-video (enableSegment false): the service treats the entire video file as a single segment and extracts metadata across its full duration. Microsoft's listed use cases are compliance checks that look for specific brand-safety issues anywhere in an ad, and full-length descriptive summaries.
- Custom segmentation (enableSegment true): you describe the segmentation logic in natural language in contentCategories, and the model creates segments to match, each with its own field values.
The agency wants one verdict and one summary per ad, which is the whole-video use case nearly word for word. Turning custom segmentation off (enableSegment false, which is also the default per the analyzer reference) gives one segment, so each field gets one value for the whole ad.
The other options, and when you'd use them
- baseAnalyzerId prebuilt-document: Base analyzers exist per content type: prebuilt-document, prebuilt-audio, prebuilt-video and prebuilt-image. A video needs prebuilt-video.
- returnDetails true: It controls whether the response includes detail such as confidence scores, bounding boxes, text spans and metadata. It doesn't merge segments.
- Whole-video only for prebuilt analyzers: The video overview says you use these modes by setting enableSegment on a custom analyzer.
Exam tip
One verdict or summary per video = enableSegment false (whole-video). Per-scene or per-story output = enableSegment true with contentCategories.
References
Last reviewed: