Product Introduction
Intelligent Search MetaInsight provides you with search services for multiple scenarios and modalities. You can analyze files stored in COS by using natural language or structured search conditions to meet your needs for managing, analyzing, and searching stored data.
Intelligent Search leverages the existing image and other data processing capabilities of Cloud Infinite to extract file features or metadata and index them into datasets, providing you with capabilities such as aggregated statistical queries of files, image content search, and face image search.
Use Limits
Intelligent Search must be bound to a COS bucket before use. You need to activate COS and bind a bucket first. Region restriction: Only buckets in the Jakarta region can be bound, and metadata indexes can only be created for COS files in the Jakarta region.
Billing Overview
The fees for Intelligent Search MetaInsight include operator processing fees, search fees, and metadata management fees. For detailed billing information, see Intelligent Search Pricing. Feature Overview
|
Manage datasets | A dataset is a collection of file metadata and features, serving as a container for file metadata indexes. After a dataset is created, Intelligent Search MetaInsight can automatically parse basic information of files in COS, perform AI analysis on media files such as images, videos, and audio, and extract tags, people, and other information to establish metadata indexes. You can also manually establish metadata indexes for files stored in COS. |
File Metadata Query | By filtering and analyzing different fields of file metadata, you can query files that meet specified conditions and list file information by specified fields and sorting methods. |
Search images by image | Search for the top N images similar to an input image in the dataset by image search. |
Semantic image search | You can search by entering a natural language description of the image content, such as "a vast sea" or "a plump strawberry". |
Face Search | You can use the face search feature to search a dataset for the top N images with faces most similar to the face in a specified image. This is suitable for scenarios such as access control gates, attendance check-in, and school dormitory management. |
Strengths
One-stop process: Massive raw unstructured data, vectorization, and vector data storage are all completed in a closed loop within MetaInsight, simplifying the production pipeline.
Automatic trigger processing: It supports batch vectorization and scalar feature extraction for existing data, and automatically triggers vectorization for newly uploaded data without manual intervention.
Data synchronization update: When the original data in COS is modified, overwritten, or deleted, the corresponding vector data and scalar feature data in MetaInsight can be synchronized for update and deletion, reducing manual maintenance costs.
Deeply optimized vector models: Tencent Cloud deeply optimized vector models are adaptable to multiple scenarios and highly efficient.
Widely adaptable scalar feature extraction model: Tencent Cloud's deeply optimized scalar feature extraction model can extract massive tags and accurate image descriptions.
Efficient hybrid search mode: By combining dense (vector) and sparse (scalar) paths and offering rich scalar filtering capabilities, it provides efficient hybrid search capabilities.
Use Cases
Mass File Statistics Query
With Intelligent Search MetaInsight, you can search for files that meet specific conditions among hundreds of millions of files, or count the number of files that meet specific conditions. For example, you can search for files containing a keyword, calculate the total size of files in a directory, or search for images containing a specific person. You can also combine simple query conditions and aggregation operations based on your business scenarios to perform data searches in more scenarios.
Data Mining
In big data scenarios, massive collected image data can be pre-classified through Intelligent Search. After data for specified scenarios is filtered out, annotation training can be performed, greatly improving data training efficiency.
Smart Album
Use the face search feature to search for the top N images in the gallery that are most similar to a specified face, with results sorted in descending order by similarity. In a smart album, you can associate recognized faces with your contacts. After the association is completed, you can directly click a face on a photo while viewing it to make calls or send SMS messages.
E-Commerce Website
By entering a product image, you can accurately find the same or similar items in the product library and quickly retrieve the corresponding product information.
Usage Process
Step 1: Creating a Dataset
A dataset is a collection composed of file metadata and features, serving as a container for storing file metadata indexes. After a dataset is created, Intelligent Search MetaInsight can automatically parse the basic information of files in COS and perform intelligent analysis on images to extract features and build metadata indexes. For example, in an e-commerce scenario, you create an e-commerce image gallery dataset, bind a bucket to the dataset through the bucket-dataset binding API, or build metadata indexes for subsequently generated e-commerce product images through the metadata index creation API. Then, using the image search feature, users can search for images in the e-commerce image gallery through natural language or similar product images. For detailed operations, see Dataset Management. Note:
When creating a dataset, you can set a search template for the dataset. A search template defines the operator operations to be performed when metadata indexes are created for the dataset. Different operators extract different types of metadata. For details, see Search Templates and Operators. Step 2: Building a Metadata Index
After creating a dataset, you can build metadata indexes for files stored in COS. With rich metadata indexes, MetaInsight provides powerful data aggregation query, analysis, and management capabilities. We provide two methods for building metadata indexes:
Method 1: Bind a COS bucket to a dataset to automatically build metadata indexes. After the bucket is bound, MetaInsight first scans the existing data in the bucket. After the existing data is indexed, MetaInsight continues to scan new files added to the bucket and builds indexes for them. For details, see Dataset Management. Step 3: Data Search