Skip to main content
Version: 1.2.0

Release Notes

1.2.0​

Note: This version of Image API uses Face SDK v3.18.0

What's New​

  • Updated installation kit and installation method. Now it takes less time to study the documentation, as well as fewer steps to fill in the configuration parameters to install the distribution.
  • Deployment and configuration of the cluster is allocated to the smc module (Self Managed Cluster). Now you need to study a smaller amount of information in the documentation (the number of dependencies has been reduced, perception has been simplified). The administrator's work path during setup has become clearer and more consistent.
  • Istallation of Image API is separated into image-api module. Now the installation is performed through a single entry point. Installation and deployment order is detailed to understandable separable steps. A separate module is easier to maintain in subsequent releases, which speeds up the process of delivering new values to customers.
  • Added support to all image-api deployments for:
    • Horizontal Pod Autoscaler — allows you not to think about how to scale, but adapts to the load that is applied to it (scales pods in accordance with server metrics) in order to ensure efficient utilization of a hardware resource;
    • nodeSelector — helps to improve stability by avoiding accidental building of a machine with inappropriate parameters by setting limits for applications in a Kubernetes cluster;
    • ingress proxy — allows you to bypass the ImageAPI frontend when embedding a service in a third-party products and solutions. For example, in the Omni Platform it acts as an engine when working with a database, but additionally allows you to use the Image API as an independent service, thus expanding the basic capabilities of the platform.
  • Added support for deploying the solution in Amazon EKS. This feature allows you to automatically scale ImageAPI services in the Amazon cloud. This reduces the need for staff involvement, allows efficient utilization of hardware resources and reduces costs for solutions that have a floating (not constant) load from end customers/services.

Improvements​

  • Updated score formula in v12 recognition methods. Now more images can be processed as good enough for further work and searches. Previously, some photos of faces, perceived by the human eye as quite high-quality, were discarded by quality thresholds.
  • Inverted confidence in the quality-liveness-anti-spoofing module of the face-detector-liveness-estimator service. Now 1 - REAL and 0 - FAKE. This corresponds to a single principle of work and evaluation of a result, and reduces confusion.

1.1.0 (17 Mar 2023)​

Note: This version of Image API uses Face SDK v3.17.0

What's New​

  • Added the ability to change a service module;
  • Added liveness-anti-spoofing module for face-detector-liveness-estimator service;
  • Added quality-liveness-anti-spoofing for face-detector-liveness-estimator service;
  • Added support of graphics card for face-detector-template-extractor service.

Improvements​

  • Improved accuracy of age-estimator, gender-estimator, mask-estimator, emotion-estimator.

1.0.1 (10 Feb 2023)​

Note: This version of Image API uses Face SDK v3.17.0

What's New​

  • Added a new field downscale_rawsamples to the configuration file deployment.cfg.When field is enabled, each sample is downscaled to the suitable size to reduce memory consumption.