Skip to main content

Intelligent Search

Intelligent Search (MetaInsight) is an AI-powered capability based on Tencent Cloud Infinite that provides intelligent content analysis and retrieval for multimedia files such as images, videos, and documents in cloud storage, helping you quickly locate and manage massive file resources.

📌 Feature Notice

If you do not see the Intelligent Search menu in the COS console, it means your account needs to enable whitelist access first. To use this feature, please submit a ticket to apply for whitelist access from the COS team.

For more details, please refer to the COS Intelligent Search Official Documentation.

Feature Introduction​

Intelligent Search integrates multiple retrieval capabilities, supporting both scalar retrieval (traditional metadata-based retrieval) and vector retrieval (AI-powered semantic retrieval), providing multi-modal hybrid search capabilities.

Core Capabilities​

Retrieval TypeDescriptionApplication Scenarios
Basic Information RetrievalMetadata-based retrieval such as filename, size, type, creation time, etc.File management, data statistics
Image RetrievalImage-to-image and text-to-image search based on image content semanticsPhoto albums, product search and recommendation
Face RetrievalRecognize facial features in images and search photos containing the same personSecurity monitoring, attendance systems, smart albums

Retrieval Methods Comparison​

FeatureScalar RetrievalVector Retrieval
PrincipleExact matching based on file attributes (metadata)Similarity matching based on AI semantic understanding
Typical UseQuery by filename, type, timeImage content search, document semantic retrieval
AdvantagesFast query speed, precise resultsIntelligent understanding, supports fuzzy matching
Use CasesQueries with known file attributesContent relevance queries

Application Scenarios​

  • Image Resource Management: Quickly locate needed image materials
  • Photo Album Applications: Classify photos by people, places, and objects
  • E-commerce Platforms: Intelligent classification and search of product images
  • Media Asset Management: Intelligent tagging and retrieval of video materials
  • Content Creation: Quickly find materials that meet requirements

Preparation​

Before using Intelligent Search, you need to enable Cloud Infinite capability and create a Dataset.

Step 1: Enable Cloud Infinite Capability​

  1. Click "Advanced Capabilities" → "Intelligent Search" on the CloudBase Cloud Storage page
  2. Or directly visit COS Console Intelligent Search Page
    • Replace {YourBucketName} with your bucket name (e.g., 36c8-static-lowcode-0gwpl9v4125156ef-1258057692)
    • Replace {YourRegion} with your region (e.g., ap-shanghai, ap-guangzhou)
  3. Follow the page prompts to enable Cloud Infinite intelligent search capability

Step 2: Create Dataset​

⚠️ Notice

Currently, only the basic information template created when enabling Cloud Infinite is supported. Cloud storage buckets do not support creating new datasets independently. This feature will be available in future updates.

  1. Log in to COS Console

  2. Select "Intelligent Search" → "Dataset Management" in the left navigation bar

  3. Click the "Create Dataset" button

  4. Configure dataset information:

    • Dataset Name: Custom dataset name
    • Dataset Description: Optional, describe the purpose of the dataset
    • Associated Bucket: Select the COS bucket corresponding to your CloudBase cloud storage
    • File Path: Specify the file path to be searched (optional all or specified prefix)

  5. Click "Confirm" to complete creation

💰 Billing Information

Intelligent Search is billed based on actual usage, including:

  • Dataset Storage: Charged by the number of files in the dataset and storage duration
  • Vector Index: Charged by the number of files with vector indexes built
  • Retrieval Requests: Charged by the number of retrieval requests

For detailed pricing, please refer to: Cloud Infinite Billing Instructions

Step 3: Wait for Metadata Index Completion​

After creating the dataset, you need to build indexes for files:

  1. In the dataset list, click the dataset name to enter the details page
  2. Click "File Index Management"
  3. Select the file types to be indexed:
    • Image Index: Supports JPG, PNG, BMP, WebP and other formats
    • Document Index: Supports PDF, Word, Excel, PPT, TXT and other formats
    • Face Index: Automatically recognizes faces in images and builds a face library
  4. Click "Start Indexing", the system will automatically analyze files and build indexes

Indexing Time: Usually completes within 1-10 minutes after task submission, depending on file quantity and size.

Intelligent Search provides two methods: console operations and API calls.

Console Retrieval​

Through the COS console visual interface, you can perform retrieval supporting two types: basic information and multi-modal hybrid retrieval.

1. Basic Information Retrieval​

Suitable for queries based on file attributes such as filename, contentType, file size, etc.

Steps:

  1. Log in to COS Console
  2. Select "Intelligent Search" → "Data Retrieval" in the left navigation bar
  3. Select retrieval scenario as "Basic Information Retrieval"
  4. Configure query conditions:
    • Select dataset
    • Set query conditions (supports multiple condition combinations):
      • Filename: Supports exact or fuzzy matching
      • Content Type (contentType): e.g., image/jpeg, application/pdf
      • File Size: Supports range queries
      • Creation Time: Supports time range queries
      • Custom Tags: Based on custom file metadata
  5. Advanced Settings (optional):
    • Set sorting method (by time, size, etc.)
    • Configure data aggregation rules
    • Set number of results returned
  6. Click "Query" to get results
  7. Support exporting results via "Export List" button or viewing export tasks in "Task Center"

Query Condition Examples:

Query RequirementQuery Condition Setting
Find all JPG imagescontentType = "image/jpeg"
Find files larger than 1MBFile size > 1048576
Find files from last 7 daysCreation time >= Current time - 7 days
Find images with "product" in filenameFilename contains "product" AND contentType starts with "image/"

Supported Operators: =, !=, >, >=, <, <=, contains, starts with, ends with, etc.

For detailed field and operator support list, please refer to: COS Field and Operator Documentation

2. Multi-modal Retrieval​

⚠️ Feature Status

Multi-modal Retrieval is not available yet. Stay tuned.

Multi-modal retrieval supports AI retrieval scenarios based on image content and face recognition, including image-to-image search, text-to-image search, and face retrieval.

Steps:

  1. Log in to COS Console
  2. Select "Intelligent Search" → "Data Retrieval" in the left navigation bar
  3. Select retrieval scenario as "Multi-modal Hybrid Retrieval"
  4. Input retrieval content:
    • Text Retrieval: Enter descriptive text, e.g., "landscape photo with blue sky and white clouds", "red dress"
    • Image Retrieval: Upload reference image or select image from bucket (supports both general image retrieval and face retrieval)
  5. Advanced Settings:
    • Return Quantity: Set maximum number of results returned (default 10, max 100)
    • Similarity Threshold: Set minimum similarity score (0-100, recommended 60-80 for image retrieval, 70-85 for face retrieval)
    • Retrieval Type: Select to search images or all
  6. Metadata Filtering (optional):
    • Further filter by metadata conditions based on retrieval results
    • Example: Only return images created in 2024
  7. Click "Query" to get results

Multi-modal Retrieval Examples:

Image-to-Image Search:

1. Upload a "beach landscape" image
2. Set similarity threshold to 70
3. System returns all photos with similar content to the image

Text-to-Image Search:

1. Enter text description: "bride in white wedding dress"
2. Set return quantity to 20
3. System returns semantically matched images

Face Retrieval:

1. Upload a clear face photo
2. Set similarity threshold to 75
3. System returns all photos containing that person
4. Can sort by time or similarity
💡 Retrieval Optimization Tips
  • Image Retrieval: Use clear images with distinct subjects, recommended similarity threshold 60-80
  • Face Retrieval: Use clear, frontal face photos as reference images, recommended similarity threshold 70-85
  • Index Configuration: Face retrieval requires face index to be configured in the dataset in advance

API Calls​

Intelligent Search supports API calls that can be integrated into your application.

📚 API Reference

For complete API usage and parameters, please refer to: Data Retrieval API Documentation

FAQ​

How long does indexing take?​

Usually completed within 1-5 minutes after file upload, depending on file size and quantity.

What file formats are supported?​

  • Images: JPG, PNG, BMP, GIF, WebP, HEIF
  • Videos: MP4, AVI, MOV, FLV and other mainstream formats

How to improve retrieval accuracy?​

  1. High-quality Images: Use clear images with distinct subjects
  2. Multi-tag Combination: Use multiple tag combinations for retrieval
  3. Reasonable Threshold: Adjust similarity threshold appropriately for image search
  4. Manual Review: Conduct manual secondary confirmation for important retrieval results

How to build indexes for existing files?​

After enabling Intelligent Search, you need to manually trigger index building for existing files:

  1. Select "Batch Build Index" in the console
  2. Select directories or files to be indexed
  3. Wait for index task to complete

How is index data stored?​

Intelligent Search index data is managed by Cloud Infinite service and does not occupy your cloud storage space. It is recommended to sync tag information to CloudBase database for complex queries.