Skip to main content

Biometric Backend Changelog

Current release: 3.1.2 (2026-05-11)

3.1.2 — 2026-05-11​

Added
  • Added support for read-only disk permissions to successfully run the inference-service;

3.1.1 — 2026-05-07​

Fixed
  • AI models are now correctly loaded according to the capability configuration.

3.1.0 — 2026-04-30​

Added
  • Added experimental endpoint: /api/deepfake/detect/v2.
  • Added configuration options to selectively load AI models into the Triton Inference container based on capability needs (face verification, PAD, deepfake detection, morphing attack detection).
  • Added experimental endpoint to check backend capabilities: /api/capabilities.
  • Added morphing attack detection capability.
  • Added configuration options in the /api/biometric endpoint to adjust thresholds for the following models: face verification, PAD, deepfake detection, and morphing attack detection. These options are optional and use the highest threshold by default.
Fixed
  • Minor bug fixes and performance improvements.
Changed
  • Removed experimental endpoint: /api/pad/v4.

3.0.0 — 2026-02-24​

Added
  • Added experimental endpoint: /api/pad/v4.
  • Added support for logging data at different log levels.
Fixed
  • Upgraded Triton version to 24.12 to support RTX 50-series GPU cards.
  • Important: The --response-cache-byte-size flag is deprecated in this Triton version. Please remove this flag from your Kubernetes configuration. Response cache is disabled by default in the new Triton version.
Changed
  • Endpoint /api/biometricsession will be deprecated.

2.5.2 — 2025-11-17​

Fixed
  • Fixed a bug to support a three-layer architecture: web-service → biometric-service → inference-service.

2.5.1 — 2025-10-21​

Added
  • Improved model versioning in the inference service.

2.5.0 — 2025-10-17​

Added
  • The api/biometric endpoint now supports processing video data captured by the Mobai SDK.
  • The api/biometric endpoint now supports configuration options for assessing the captured live face image.
  • Implemented 8 lighting-condition assessment metrics in accordance with ISO/IEC 29794-5.
  • Upgraded the deepfake detection model to v2.
Fixed
  • Fixed issues in the biometric service video-handling process.
  • Resolved an issue in the inference service.
  • Resolved an issue with the obfuscated biometric service.

2.4.0 — 2025-05-08​

Added
  • New endpoint for additional security checks.
  • The HTTP server can now be started from the biometric service container (web service is no longer required).

2.3.2 — 2024-11-23​

Fixed
  • Issue where images captured from the biometric capture SDK could generate incorrect detection error codes.

2.3.1 — 2024-09-13​

Changed
  • Reverted to a previous revision of the inference service base image to address processing speed degradation.

2.3.0 — 2024-08-29​

Added
  • Support for multiple versions of capture constraints in the mobile capture SDK.
Changed
  • Improved presentation attack detection performance.
Fixed
  • Fixed an issue where “no face detected” caused an error.

Older changelog entries have been archived to keep this page focused on recent releases.