Releases: IntelLabs/vdms
Release 2.10.0
What's New
Updated Docker Image
- Docker image is available on Dockerhub
Improvements to Descriptor/Descriptor Set
- Cache DescriptorSet path to bypass database query after initial setup
- Additional capability for batch inserts of descriptors
- Add Faiss Hierarchical Navigable Small World (HNSW) Engine for indexing and searching descriptors
- Multiple optimizations within the FindDescriptor path, improving performance by up to 200x depending on the specifics on the query
Bug Fixes
- Connections for Neo4J are now being returned and re-used after transaction commit
- Removed unique constraint for link based descriptor queries
Updated Dependencies
Updated documentation in Wiki
Release 2.9.0
What's New
Updated Docker Image
- Docker image is available on Dockerhub
New mutual TLS support for C++ Client
New Timing Capability (Experimental)
- The new timing capability provides ability to output a variety of timing information for use in tuning and optimization.
- Additional details are available: Timing Capability
Dynamic Metadata
- VDMS can now dynamically add metadata from images/videos by leveraging the UDF capability.
- Additional details are available: Dynamic Metadata
Improved Neo4j Handler Error Handling and Reporting
Bug Fixes
Updated Dependencies
Updated documentation in Wiki
Release 2.8.0
What's New
Updated Docker Image
- Docker image is available on Dockerhub
- Users have the ability to override the default parameters in config-vdms.json. See Environment Variable in Docker Container for more information
New Neo4J + S3 Query Handler (Experimental)
- The new query handler takes advantage of our S3 connector and adds Neo4J for metadata store instead of default PMGD.
- Additional details are available: Neo4J Query Handler, NeoAdd, NeoFind
New mutual TLS support
- mTLS support was added to the server and Python client while retaining support for running in a TLS only (transport security only, no client authentication) or an insecure mode
- Additional details are available: VDMS Mutual TLS Support
Add Entity via File Path
- VDMS can now point to an image/video file accessible by the filesystem instead of storing the blob in the data store. The file could be on the local filesystem or a remote filesystem mounted on the server running VDMS.
- Additional details are available: Updated AddImage Wiki, Updated AddVideo Wiki
Bug Fixes
Updated documentation in Wiki
Release 2.7.0
What's New
Updated Docker Image
- Docker image is available on Dockerhub
New FindDescriptorSet Command (Experimental)
- The new command provides capability to obtain information regarding a descriptor set and to explicitly perform a store index operation.
- Additional details are available: FindDescriptorSet Command
Configurable S3 Connection
- New parameters accepted in VDMS configuration file for S3 storage address, proxy details, and AWS logging control.
- Additional details are available: ConfigFile
Bug Fixes
Updated documentation in Wiki
Release 2.6.0
What's New
Updated Docker Image
- Docker image is available on Dockerhub
New Query Handler Selector (Experimental)
- Enhancement to provide developers the ability to either modify existing query handlers (e.g. changing data and operations flow) or develop entirely new handlers from scratch (e.g. enabling a new type of metadata backend).
- Additional details are available; Query Handler Development
Redesign of Video Class (Experimental)
- Experimental enhancement to video management design to work with videos as a single entity and ensures the Out-of-Memory exceptions are caught/addressed
Upgrade to Protobuf v24.2
- Installation instructions updated in Install.md and Wiki
Improved Error Handling
- Improve how errors/exceptions are handled and provide better information to the client
Bug Fixes
Updated documentation in Wiki
Release 2.5.0
What's New
-
New Debian-based Dockerfile
- Docker image is available on Dockerhub
-
New User-defined Operations (UDF) and Remote Operations
- Users can apply user-defined operations to images and execute image operations on a remote server using Flask APIs
- Additional details are available; User Defined Operations and Remote Operations
-
AWS S3 Support
- Users can write data to AWS S3 instead of using local storage by specifying
storage_type
in configuration file - Additional details are available: ConfigFile Wiki
- Users can write data to AWS S3 instead of using local storage by specifying
-
Upgrade to TileDB 2.14
-
Updated documentation in Wiki
Release 2.4.0
What's New
-
Support for Storing and Accessing Other Data Formats
- Users can use the new AddBlob, FindBlob, and UpdateBlob commands to add and query other data formats.
- Additional details are available: AddBlob, FindBlob, and UpdateBlob.
-
CSV Parser for C++ Client
- Users can ingest data from a CSV file instead of writing JSON queries. This feature supports the Insert Commands.
- Additional details are available: CSV Plugin for C++ Client
-
Upgraded dependencies
-
Updated documentation in Wiki
Release V2.3.0
What's New
-
Filters to Identify Near-Neighbor Groups (FLINNG):
- Integrated a new indexing and similarity search library (FLINNG) that addresses the computational challenges of high-dimensional similarity search by presenting an index with fast construction time, low memory requirement, and zero query-time distance computations
- Instructions are provided (https://github.com/IntelLabs/vdms/wiki/FLINNG-Library-and-Performance-Details)
-
Periodic backup (Auto-replication):
- Users can automatically backup db folder of VDMS periodically and restore the database as needed
- Instructions are provided (https://github.com/IntelLabs/vdms/wiki/VDMS-Backup-and-Restoring-Options)
-
Kafka Connectors:
- We are moving towards distributed VDMS and have provided Kafka connectors, configurations, instructions, and examples to use VDMS in a distributed manner leveraging Kafka
- Instructions are provided (https://github.com/IntelLabs/vdms/wiki/Run-Kafka-VDMS-Example)
-
Video Improvement
- Addressed memory leak in video command
-
Independent C++ Client:
- Ability to build client independently without building VDMS server
Release 2.2.0
What's New
- Flexible Retention Policy (Auto-deletion):
- Users can automatically have node removed from VDMS
- Internal Database Retry:
- Retry query before asserting and exiting database
- Allow User-defined Operators in VCL:
- User can define a new operator in VCL and expose it in the VDMS API. Provided a sample custom operation
- Improve Descriptor Search:
- Fixed descriptor bug (insert multiple descriptors)
- Added the ability to search descriptors by query blob AND constraints
- JPEG Storage Quality:
- Image differences due to lossy compression of JPEG. Added disclaimer to user to use lossless format (PNG) for inference applications (https://github.com/IntelLabs/vdms/wiki/AddImage#inference-image-format )
- Increase Buffer Size:
- We increased the max buffer size for database transactions to 1GB and set a default size of 32MB (previous max)
- Removed Dockerfiles:
- Only provide base Dockerfile
2021_08_12 Release
This release includes an updated version of VDMS that compiles against updated versions of OpenCV, FAISS, and Protobuf.
Features Added:
- Data deletion and expiration
- VCL Image storage to Binary