Skip to content

v.5.147.0#

Changes

  • SDK has been updated to version 5.35.0.

    In this version:

    • Photorealistic Face estimator has been updated to version 1;
    • Face Image Modification estimator has been updated to version 2.

  • New checks have been added to the face_quality check group for the "/handlers" and "/verifiers" resources:

    • neck_occlusion — determines whether the neck is occluded by clothing or accessories. For more information, see the description of the Neck Occlusion estimator.
    • photorealistic_face — determines whether the face in the image is real (photographic) or drown. For more information, see the description of the Photorealistic Face estimator.
    • duty_uniform — determines whether uniform clothing (epaulettes) is present on the upper body. For more information, see the Duty Uniform estimator description.
    • clothes_color_lightness — determines whether clothing on the upper body (shoulder area) is light or dark. For more information, see the Clothes Color Lightness estimator description.

  • New arguments have been added to the Remote SDK container launch command to enable/disable the use of new estimators: --enable-duty-uniform-estimator, --enable-neck-occlusion-estimator, --enable-photorealistic-face-estimator, and --enable-light-colored-clothes-estimator.

    For more information see Enabling/Disabling several estimators and detectors section.

  • The Remote SDK service now supports a new type of plugin that allows you to replace built-in estimators with custom implementations (custom neural networks or alternative algorithms), as well as optimize existing estimators without changing the service's source code.

    For more information, see the sections of the Remote SDK service developer's manual:

  • For the Remote SDK service, the ability to switch versions of the Liveness and Deepfake algorithms has been added via settings in the Configurator.

    Previously, changing the algorithm version required manually editing the configuration file inside the container. Now version management has been moved to the Configurator settings: the version parameter has been added to the LUNA_REMOTE_SDK_LIVENESS_ESTIMATOR_SETTINGS and LUNA_REMOTE_SDK_DEEPFAKE_ESTIMATOR_SETTINGS parameter groups. This parameter accepts integer values: 0 — the default algorithm version is used.

    This simplifies configuration and testing of different algorithm versions. For more information, see Managing Liveness and Deepfake neural network versions section.

  • Now LUNA PLATFORM supports images in the WebP, AVIF, HEIC, and GIF formats.

    Important! These formats support animation. When sending an animated image, only the first frame is used for processing. The remaining frames are ignored. This ensures correct processing of both static and animated images.

    For more details, see /images, /events, /sdk, /iso and /verifiers.

  • In the Admin, Faces, Handlers and Events services, support for obsolete encodings has been removed: ISO-8859-x and KOI8-R/U.

  • A new query parameter depth has been added to the get list of folder images and get list of folder objects requests.

    This parameter controls the nesting level when retrieving a list of objects from a folder:

    • depth=0 (default value): all objects within the folder are returned, including the contents of subfolders (recursively). This matches the behavior in previous versions of LUNA PLATFORM.
    • depth=1: only objects and subfolders of the first nesting level are returned. Subfolder names are returned with a trailing slash, while file names are returned without. This allows you to retrieve the contents of a folder without recursively loading all objects, which can improve response times when working with deep hierarchies.

  • The Image Store service documentation section has been updated.

    This section has been expanded and structured to reflect the service's current capabilities. Descriptions of storage types, object storage features, folder management (restrictions, access rights, examples for local and S3 storage), and clarified TTL configuration rules with priorities (object → folder → bucket) have been added.

    See "Image Store service".

  • The documentation section "Stream format supported by Video Agent service" has been updated to include information about CPU and GPU decoding features.

  • A new query parameter check_agent has been added to the create stream and put stream requests.

    This parameter controls the availability check for agents to process the analytics specified in the request:

    • check_agent=1 (default): If no agents are available, the stream will not be created, and the request will fail;
    • check_agent=0: The stream is created even if no agents are available and will remain in the pending state until they become available. This allows streams to be created without waiting for agents to recover after restarts or short-term outages.

    Note: If a stream is created but no available agents have free slots, it will also remain in the pending status until resources are freed (regardless of the check_agent value).

  • The set of collected metrics for Agent-lambda has been expanded.

    A new value, eof_stream, has been added to the agent_counter_total counter metric for the agent_action label. This allows tracking instances where streams terminate upon reaching end-of-file, complementing the existing stream lifecycle and frame processing metrics.

Fixed errors

  • Fixed a bug due to which the custom image (base_image) specified when creating, updating or importing lambdas of type agent, was replaced with the default image (lpa-lambda-base).