|
| 1 | +--- |
| 2 | +title: People's featured clothing |
| 3 | +description: This article gives an overview of featured clothing images appearing in a video. |
| 4 | +ms.topic: conceptual |
| 5 | +ms.date: 11/15/2021 |
| 6 | +ms.author: juliako |
| 7 | +--- |
| 8 | + |
| 9 | +# People's featured clothing (preview) |
| 10 | + |
| 11 | +Azure Video Indexer enables you to get data on the featured clothing of an observed person. The people's featured clothing feature, helps to enable the following scenarios: |
| 12 | + |
| 13 | +- Ads placement - using the featured clothing insight information, you can enable more targeted ads placement. |
| 14 | +- Video summarization - you can create a summary of the most interesting outfits appearing in the video. |
| 15 | + |
| 16 | +## Viewing featured clothing |
| 17 | + |
| 18 | +The featured clothing insight is available when indexing your file by choosing the Advanced option -> Advanced video or Advanced video + audio preset (under Video + audio indexing). Standard indexing will not include this insight. |
| 19 | + |
| 20 | +:::image type="content" source="./media/detected-clothing/index-video.png" alt-text="This screenshot represents an indexing video option."::: |
| 21 | + |
| 22 | +The featured clothing images are ranked based on some of the following factors: key moments of the video, general emotions from text or audio. The `id` property indicates the ranking index. For example, `"id": 1` signifies the most important featured clothing. |
| 23 | + |
| 24 | +> [!NOTE] |
| 25 | +> The featured clothing currently can be viewed only from the artifact file. |
| 26 | +
|
| 27 | +1. In the right-upper corner, select to download the artifact zip file: **Download** -> **Artifact (ZIP)** |
| 28 | +1. Open `featuredclothing.zip`. |
| 29 | + |
| 30 | +The .zip file contains two objects: |
| 31 | + |
| 32 | +- `featuredclothing.map.json` - the file contains instances of each featured clothing, with the following properties: |
| 33 | + |
| 34 | + - `id` – ranking index (`"id": 1` is the most important clothing). |
| 35 | + - `confidence` – the score of the featured clothing. |
| 36 | + - `frameIndex` – the best frame of the clothing. |
| 37 | + - `timestamp` – corresponding to the frameIndex. |
| 38 | + - `opBoundingBox` – bounding box of the person. |
| 39 | + - `faceBoundingBox` – bounding box of the person's face, if detected. |
| 40 | + - `fileName` – where the best frame of the clothing is saved. |
| 41 | + |
| 42 | + An example of the featured clothing with `"id": 1`. |
| 43 | + |
| 44 | + ``` |
| 45 | + "instances": [ |
| 46 | + { |
| 47 | + "confidence": 0.98, |
| 48 | + "faceBoundingBox": { |
| 49 | + "x": 0.50158, |
| 50 | + "y": 0.10508, |
| 51 | + "width": 0.13589, |
| 52 | + "height": 0.45372 |
| 53 | + }, |
| 54 | + "fileName": "frame_12147.jpg", |
| 55 | + "frameIndex": 12147, |
| 56 | + "id": 1, |
| 57 | + "opBoundingBox": { |
| 58 | + "x": 0.34141, |
| 59 | + "y": 0.16667, |
| 60 | + "width": 0.28125, |
| 61 | + "height": 0.82083 |
| 62 | + }, |
| 63 | + "timestamp": "00:08:26.6311250" |
| 64 | + }, |
| 65 | + ``` |
| 66 | +- `featuredclothing.frames.map` – this folder contains images of the best frames that the featured clothing appeared in, corresponding to the `fileName` property in each instance in `featuredclothing.map.json`. |
| 67 | +
|
| 68 | +## Limitations and assumptions |
| 69 | +
|
| 70 | +It's important to note the limitations of featured clothing to avoid or mitigate the effects of false detections of images with low quality or low relevancy. |
| 71 | +
|
| 72 | +- Pre-condition for the featured clothing is that the person wearing the clothes can be found in the observed people insight. |
| 73 | +- If the face of a person wearing the featured clothing wasn't detected, the results won't include the faces bounding box. |
| 74 | +- If a person in a video wears more than one outfit, the algorithm selects its best outfit as a single featured clothing image. |
| 75 | +- When posed, the tracks are optimized to handle observed people who most often appear on the front. |
| 76 | +- Wrong detections may occur when people are overlapping. |
| 77 | +- Frames containing blurred people are more prone to low quality results. |
| 78 | +
|
| 79 | +For more information, see the [limitations of observed people](observed-people-tracing.md#limitations-and-assumptions). |
| 80 | +
|
| 81 | +## Next steps |
| 82 | +
|
| 83 | +- [Trace observed people in a video](observed-people-tracing.md) |
| 84 | +- [People's detected clothing](detected-clothing.md) |
0 commit comments