Edit

Share via

Do video retrieval using vectorization (version 4.0 preview)

  • Article

Important

On 30 June 2025, Azure AI Vision Video Retrieval will be retired. The decision to retire this feature is part of our ongoing effort to improve and simplify and improve the features offered for video processing. Migrate to Azure AI Content Understanding and Azure AI Search to benefit from their additional capabilities.

Video processing: Video Retrieval vs Azure AI Content Understanding

Feature Video Retrieval for video description Azure AI Content Understanding
Video Length Supported Optimized for short videos, up to ~3 minutes Supports short & long videos, up to 4 hours
Frame Processing Up to 20 frames Batch processing, sampling shot-by-shot sampled across entire video
Content Extraction Pre-Processing Transcription Transcription, Shot identification, Face grouping
Structured Output Support Not supported Supports schema-conforming structured outputs
Data types Video supported Video, images, documents, and speech supported
Pricing Variable Token-based Fixed cost per minute of video processed

To migrate to Content Understanding for video summaries and descriptions, we'd recommend reviewing the Azure AI Content Understanding documentation.

Video Search: Video Retrieval vs. Azure AI Search and Content Understanding

Feature Video Retrieval for video search Azure AI Search and Content Understanding
Visual Embedding type Frame-based Image Embeddings Video description text embeddings
Content Extraction Pre-Processing Transcription, OCR Transcription, Shot identification, Face grouping
People & Object search support Strong support Strong support
Action and Event support Limited Strong support
Customization None Content Understanding analyzer can be customized to focus using the fields and field descriptions

To start building the search use case with Content Understanding, we recommend starting with this sample which shows how to use Azure AI Search to search videos.

To avoid service disruptions, migrate by 30 June 2025.

Azure AI Video Retrieval APIs are part of Azure AI Vision and enable developers to create a media index, add documents (videos and images) to it, and search it with natural language. Developers can define metadata schemas for each index and ingest metadata to the service to help with retrieval. Developers can also specify which features to extract from the index (vision, speech) and filter their search based on features.

Prerequisites

Select a video

Refer to the video input requirements:

Supported formats

File format Description
asf ASF (Advanced / Active Streaming Format)
avi AVI (Audio Video Interleaved)
flv FLV (Flash Video)
matroskamm, webm Matroska / WebM
mov,mp4,m4a,3gp,3g2,mj2 QuickTime / MOV

Supported video codecs

Codec Format
h264 H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10
h265 H.265/HEVC
libvpx-vp9 libvpx VP9 (codec vp9)
mpeg4 MPEG-4 part 2

Supported audio codecs

Codec Format
aac AAC (Advanced Audio Coding)
mp3 MP3 (MPEG audio layer 3)
pcm PCM (uncompressed)
vorbis Vorbis
wmav2 Windows Media Audio 2

Call the Video Retrieval APIs

To use the Video Retrieval APIs in a typical pattern, you'll follow these steps:

  1. Create an index using PUT - Create an index.
  2. Add video documents to the index using PUT - CreateIngestion.
  3. Wait for the ingestion to complete, checking with GET - ListIngestions.
  4. Search for a keyword or phrase using POST - SearchByText.

The Video Retrieval APIs allow a user to add metadata to video files. Metadata is additional information associated with video files such as "Camera ID," "Timestamp," or "Location" that can be used to organize, filter, and search for specific videos. This example demonstrates how to create an index, add video files with associated metadata, and perform searches using different features.

Step 1: Create an Index

To begin, you need to create an index to store and organize the video files and their metadata. The example below demonstrates how to create an index named "my-video-index" using the Create Index API.

curl.exe -v -X PUT "https://<YOUR_ENDPOINT_URL>/computervision/retrieval/indexes/my-video-index?api-version=2023-05-01-preview" -H "Ocp-Apim-Subscription-Key: <YOUR_SUBSCRIPTION_KEY>" -H "Content-Type: application/json" --data-ascii "
{
  'metadataSchema': {
    'fields': [
      {
        'name': 'cameraId',
        'searchable': false,
        'filterable': true,
        'type': 'string'
      },
      {
        'name': 'timestamp',
        'searchable': false,
        'filterable': true,
        'type': 'datetime'
      }
    ]
  },
  'features': [
    {
      'name': 'vision',
      'domain': 'surveillance'
    },
    {
      'name': 'speech'
    }
  ]
}"

Response:

HTTP/1.1 201 Created
Content-Length: 530
Content-Type: application/json; charset=utf-8
request-id: cb036529-d1cf-4b44-a1ef-0a4e9fc62885
api-supported-versions: 2023-01-15-preview,2023-05-01-preview
x-envoy-upstream-service-time: 202
Date: Thu, 06 Jul 2023 18:05:05 GMT
Connection: close

{
  "name": "my-video-index",
  "metadataSchema": {
    "language": "en",
    "fields": [
      {
        "name": "cameraid",
        "searchable": false,
        "filterable": true,
        "type": "string"
      },
      {
        "name": "timestamp",
        "searchable": false,
        "filterable": true,
        "type": "datetime"
      }
    ]
  },
  "userData": {},
  "features": [
    {
      "name": "vision",
      "modelVersion": "2023-05-31",
      "domain": "surveillance"
    },
    {
      "name": "speech",
      "modelVersion": "2023-06-30",
      "domain": "generic"
    }
  ],
  "eTag": "\"7966244a79384cca9880d67a4daa9eb1\"",
  "createdDateTime": "2023-07-06T18:05:06.7582534Z",
  "lastModifiedDateTime": "2023-07-06T18:05:06.7582534Z"
}

Step 2: Add video files to the index

Next, you can add video files to the index with their associated metadata. The example below demonstrates how to add two video files to the index using SAS URLs with the Create Ingestion API.

curl.exe -v -X PUT "https://<YOUR_ENDPOINT_URL>/computervision/retrieval/indexes/my-video-index/ingestions/my-ingestion?api-version=2023-05-01-preview" -H "Ocp-Apim-Subscription-Key: <YOUR_SUBSCRIPTION_KEY>" -H "Content-Type: application/json" --data-ascii "
{
  'videos': [
    {
      'mode': 'add',
      'documentId': '02a504c9cd28296a8b74394ed7488045',
      'documentUrl': 'https://example.blob.core.windows.net/videos/02a504c9cd28296a8b74394ed7488045.mp4?sas_token_here',
      'metadata': {
        'cameraId': 'camera1',
        'timestamp': '2023-06-30 17:40:33'
      }
    },
    {
      'mode': 'add',
      'documentId': '043ad56daad86cdaa6e493aa11ebdab3',
      'documentUrl': '[https://example.blob.core.windows.net/videos/043ad56daad86cdaa6e493aa11ebdab3.mp4?sas_token_here',
      'metadata': {
        'cameraId': 'camera2'
      }
    }
  ]
}"

Response:

HTTP/1.1 202 Accepted
Content-Length: 152
Content-Type: application/json; charset=utf-8
request-id: ee5e48df-13f8-4a87-a337-026947144321
operation-location: http://api.example.com.trafficmanager.net/retrieval/indexes/my-test-index/ingestions/my-ingestion
api-supported-versions: 2023-01-15-preview,2023-05-01-preview
x-envoy-upstream-service-time: 709
Date: Thu, 06 Jul 2023 18:15:34 GMT
Connection: close

{
  "name": "my-ingestion",
  "state": "Running",
  "createdDateTime": "2023-07-06T18:15:33.8105687Z",
  "lastModifiedDateTime": "2023-07-06T18:15:34.3418564Z"
}

Step 3: Wait for ingestion to complete

After you add video files to the index, the ingestion process starts. It might take some time depending on the size and number of files. To ensure the ingestion is complete before performing searches, you can use the Get Ingestion API to check the status. Wait for this call to return "state" = "Completed" before proceeding to the next step.

curl.exe -v -X GET "https://<YOUR_ENDPOINT_URL>/computervision/retrieval/indexes/my-video-index/ingestions?api-version=2023-05-01-preview&$top=20" -H "ocp-apim-subscription-key: <YOUR_SUBSCRIPTION_KEY>"

Response:

HTTP/1.1 200 OK
Content-Length: 164
Content-Type: application/json; charset=utf-8
request-id: 4907feaf-88f1-4009-a1a5-ad366f04ee31
api-supported-versions: 2023-01-15-preview,2023-05-01-preview
x-envoy-upstream-service-time: 12
Date: Thu, 06 Jul 2023 18:17:47 GMT
Connection: close

{
  "value": [
    {
      "name": "my-ingestion",
      "state": "Completed",
      "createdDateTime": "2023-07-06T18:15:33.8105687Z",
      "lastModifiedDateTime": "2023-07-06T18:15:34.3418564Z"
    }
  ]
}

Step 4: Perform searches with metadata

After you add video files to the index, you can search for specific videos using metadata. This example demonstrates two types of searches: one using the "vision" feature and another using the "speech" feature.

Search with "vision" feature

To perform a search using the "vision" feature, use the Search By Text API with the vision filter, specifying the query text and any other desired filters.

curl.exe -v -X POST "https://<YOUR_ENDPOINT_URL>/computervision/retrieval/indexes/my-video-index:queryByText?api-version=2023-05-01-preview" -H "Ocp-Apim-Subscription-Key: <YOUR_SUBSCRIPTION_KEY>" -H "Content-Type: application/json" --data-ascii "
{
  'queryText': 'a man with black hoodie',
  'filters': {
    'stringFilters': [
      {
        'fieldName': 'cameraId',
        'values': [
          'camera1'
        ]
      }
    ],
    'featureFilters': ['vision']
  }
}"

Response:

HTTP/1.1 200 OK
Content-Length: 3289
Content-Type: application/json; charset=utf-8
request-id: 4c2477df-d89d-4a98-b433-611083324a3f
api-supported-versions: 2023-05-01-preview
x-envoy-upstream-service-time: 233
Date: Thu, 06 Jul 2023 18:42:08 GMT
Connection: close

{
  "value": [
    {
      "documentId": "02a504c9cd28296a8b74394ed7488045",
      "documentKind": "VideoFrame",
      "start": "00:01:58",
      "end": "00:02:09",
      "best": "00:02:03",
      "relevance": 0.23974405229091644
    },
    {
      "documentId": "02a504c9cd28296a8b74394ed7488045",
      "documentKind": "VideoFrame",
      "start": "00:02:27",
      "end": "00:02:29",
      "best": "00:02:27",
      "relevance": 0.23762696981430054
    },
    {
      "documentId": "02a504c9cd28296a8b74394ed7488045",
      "documentKind": "VideoFrame",
      "start": "00:00:26",
      "end": "00:00:27",
      "best": "00:00:26",
      "relevance": 0.23250913619995117
    },
  ]
}

Search with "speech" feature

To perform a search using the "speech" feature, use the Search By Text API with the speech filter, providing the query text and any other desired filters.

curl.exe -v -X POST "https://<YOUR_ENDPOINT_URL>com/computervision/retrieval/indexes/my-video-index:queryByText?api-version=2023-05-01-preview" -H "Ocp-Apim-Subscription-Key: <YOUR_SUBSCRIPTION_KEY>" -H "Content-Type: application/json" --data-ascii "
{
  'queryText': 'leave the area',
  'dedup': false,
  'filters': {
    'stringFilters': [
      {
        'fieldName': 'cameraId',
        'values': [
          'camera1'
        ]
      }
    ],
    'featureFilters': ['speech']
  }
}"

Response:

HTTP/1.1 200 OK
Content-Length: 49001
Content-Type: application/json; charset=utf-8
request-id: b54577bb-1f46-44d8-9a91-c9326df3ac23
api-supported-versions: 2023-05-01-preview
x-envoy-upstream-service-time: 148
Date: Thu, 06 Jul 2023 18:43:07 GMT
Connection: close

{
  "value": [
    {
      "documentId": "02a504c9cd28296a8b74394ed7488045",
      "documentKind": "SpeechTextSegment",
      "start": "00:07:07.8400000",
      "end": "00:07:08.4400000",
      "best": "00:07:07.8400000",
      "relevance": 0.8597901463508606
    },
    {
      "documentId": "02a504c9cd28296a8b74394ed7488045",
      "documentKind": "SpeechTextSegment",
      "start": "00:07:02.0400000",
      "end": "00:07:03.0400000",
      "best": "00:07:02.0400000",
      "relevance": 0.8506758213043213
    },
    {
      "documentId": "02a504c9cd28296a8b74394ed7488045",
      "documentKind": "SpeechTextSegment",
      "start": "00:07:10.4400000",
      "end": "00:07:11.5200000",
      "best": "00:07:10.4400000",
      "relevance": 0.8474636673927307
    }
  ]
}

Next step

Multimodal embeddings concepts