The Video Embedding extractor processes video content and generates vector embeddings that enable semantic search capabilities.

​
Overview

The Video Embedding feature extractor analyzes video content to generate vector representations (embeddings) that capture the semantic meaning of the visual content. These embeddings enable powerful semantic search capabilities across video libraries.

​
Required Inputs

ParameterTypeRequiredDefaultDescription
video_urlstringYes-URL pointing to the video file to be processed. Supported formats: MP4, MOV, AVI
model_namestringNo”default”Name of the embedding model to use
embedding_intervalfloatNo5.0Interval (in seconds) at which to extract embeddings
include_audiobooleanNotrueWhether to include audio features in the embedding

​
Configurations

​
Embedding Models

The extractor supports different embedding models optimized for various use cases:

ModelDescriptionBest For
defaultBalanced model for general contentGeneral video search
visual-detailedFocuses on detailed visual featuresVisual content search
multimodalCombines visual and audio signalsRich multimedia content

​
Configuration Examples

Default
{
  "model_name": "default",
  "embedding_dimension": 512,
  "normalize": true
}
Visual Detailed
{
  "model_name": "visual-detailed",
  "embedding_dimension": 1024,
  "normalize": true,
  "include_audio": false
}
Multimodal
{
  "model_name": "multimodal",
  "embedding_dimension": 768,
  "normalize": true,
  "include_audio": true,
  "audio_weight": 0.4
}

​
Base Configuration

OptionTypeDefaultDescription
embedding_intervalfloat5.0Time interval (in seconds) between embedding extractions
normalizebooleantrueWhether to normalize embedding vectors
embedding_dimensioninteger512Dimension of the output embedding vector

​
Configuration Examples

Sample
{
  "embedding_interval": 2.5,    // Extract embeddings every 2.5 seconds
  "normalize": true,            // Normalize embedding vectors
  "embedding_dimension": 512    // 512-dimensional embeddings
}

​
Processing Flow

​
Output Schema

This feature extractor will output as a feature vector in the feature store.

{
  "document_id": "doc_abc123",
  "collection_id": "col_xyz789",
  "source_object_id": "obj_def456",
  
  // Embedding metadata
  "embedding": {
    "model": "default",
    "dimension": 512,
    "timestamp": 15.5,         // Timestamp in seconds
    "vector": [0.23, 0.45, ...], // Truncated for brevity
    "normalized": true
  },
  
  // Video metadata
  "video": {
    "filename": "product_demo.mp4",
    "timestamp": 15.5,         // Timestamp in seconds
    "total_duration": 325.7,   // Total video duration in seconds
    "width": 1920,
    "height": 1080,
    "fps": 30
  }
}

The Video Embedding extractor processes video content and generates vector embeddings that enable semantic search capabilities.

​
Overview

The Video Embedding feature extractor analyzes video content to generate vector representations (embeddings) that capture the semantic meaning of the visual content. These embeddings enable powerful semantic search capabilities across video libraries.

​
Required Inputs

ParameterTypeRequiredDefaultDescription
video_urlstringYes-URL pointing to the video file to be processed. Supported formats: MP4, MOV, AVI
model_namestringNo”default”Name of the embedding model to use
embedding_intervalfloatNo5.0Interval (in seconds) at which to extract embeddings
include_audiobooleanNotrueWhether to include audio features in the embedding

​
Configurations

​
Embedding Models

The extractor supports different embedding models optimized for various use cases:

ModelDescriptionBest For
defaultBalanced model for general contentGeneral video search
visual-detailedFocuses on detailed visual featuresVisual content search
multimodalCombines visual and audio signalsRich multimedia content

​
Configuration Examples

Default
{
  "model_name": "default",
  "embedding_dimension": 512,
  "normalize": true
}
Visual Detailed
{
  "model_name": "visual-detailed",
  "embedding_dimension": 1024,
  "normalize": true,
  "include_audio": false
}
Multimodal
{
  "model_name": "multimodal",
  "embedding_dimension": 768,
  "normalize": true,
  "include_audio": true,
  "audio_weight": 0.4
}

​
Base Configuration

OptionTypeDefaultDescription
embedding_intervalfloat5.0Time interval (in seconds) between embedding extractions
normalizebooleantrueWhether to normalize embedding vectors
embedding_dimensioninteger512Dimension of the output embedding vector

​
Configuration Examples

Sample
{
  "embedding_interval": 2.5,    // Extract embeddings every 2.5 seconds
  "normalize": true,            // Normalize embedding vectors
  "embedding_dimension": 512    // 512-dimensional embeddings
}

​
Processing Flow

​
Output Schema

This feature extractor will output as a feature vector in the feature store.

{
  "document_id": "doc_abc123",
  "collection_id": "col_xyz789",
  "source_object_id": "obj_def456",
  
  // Embedding metadata
  "embedding": {
    "model": "default",
    "dimension": 512,
    "timestamp": 15.5,         // Timestamp in seconds
    "vector": [0.23, 0.45, ...], // Truncated for brevity
    "normalized": true
  },
  
  // Video metadata
  "video": {
    "filename": "product_demo.mp4",
    "timestamp": 15.5,         // Timestamp in seconds
    "total_duration": 325.7,   // Total video duration in seconds
    "width": 1920,
    "height": 1080,
    "fps": 30
  }
}