What's New: release notes
Documentation version: v15
Last updated
Documentation version: v15
Last updated
5.6.3 - 5.6.1 - 2022-10-06Fixed
Distance detection on tablets and desktops was improved
Countdown settings were extended and added to API. See more in
Several minor issues are fixed.
5.6.0 - 2022-08-05New
Events status added. Starting from this version, there is a new observable events$, where LIQA returns information about its status and errors.
LIQA informs about exact loading stages (allows to make progress bar)
LIQA informs about successful loading finish and successful camera start.
LIQA informs about issues like "camera access not granted" or general errors, that occurred during loading or execution.
Learn more about supported events here:
See examples of events parsing here: Events parsing
Fixed
Several minor issues are fixed.
5.5.1 - 5.5.0 - 2022-06-15Fixed
We changed the way how LIQA treats provided canvas for video / requested size of canvas. previously, LIQA attempted to set up a videocapture from device camera to the desired canvas size. In case if device camera was not able to provide such resolution or the aspect ratio was incompatible with camera, the video either did not start (desktops, laptops) or started with very skewed face image (tablets, mostly Apple iPad).
New version fixes this: LIQA adjusts the requested canvas size to the available camera size and centers the canvas with video. This allows LIQA start correctly on nearly any device with somewhat like responsive UI.
5.4.0 - 2022-03-31We added a new feature to check final image, captured by LIQA!
Previously, LIQA performed analysis of real time video frames, but after approving the "photo capturing" action, the final image was not processed. This caused the situation that some users took pictures of not propper quality and LIQA had not validated it.
Fixed
We improved protection of API methods from incorrect input type. Previously it could cause an issue with constant "too close" output in qualityStatus$.face_position attribute.
We updated internal dependencies on FaceDetection models. This allowed to reduce time for one frame processing in real time video. The improvement is mostly observed on some devices of 2018 production year.
iOS 15.3 has a vulnerability in Safary web browser. The issue prevents LIQA from correct behaviour. The issue result in case that calling .play() method of LIQA 2nd time result in failing video frames processing.
5.3.0 - 2022-02-22Fixed
Improved WebAssembly versioning control. Now you will see a console message for checking the compliance of wasm scripts with a loaded version of the library.
5.2.0 - 2021-12-07Fixed
Fixed several bugs with face rotation logic
Fixed several bugs with fetching models and WASM interactions during load method
Now load and play methods correctly reject messages
5.1.1 - 5.1.4 - 2021-09-30Fixed
Fixed bugs with camera on iOS devices and distance levels
Added iOS 15.0 support
5.1.0 - 2021-09-17Fixed
Reworked head rotation estimation model. Improved head angles tracking on different distances from camera.
Reworked distance estimation model. Fixed issue with incorrect distance calculation for head rotated to the left / to the right.
Reworked camera user agents: now it is specific for Android and iOS and allows the correct aspect ratio settings for different devices.
Fixed behaviour for devices with multiple cameras available.
Modified settings for distance tracking, added new parameter in config. Added new output for facePosition observable output.
5.0.7 - 5.0.8 - 2021-09-10Fixed
Incorrect aspect ratio of final image - it was different from live videostream canvas size on some devices.
5.0.3 - 5.0.6 - 2021-08-30Fixed
Retake selfie action sometimes stuck with no camera re-initialized
Fixed incorrect symbols position in AutoSelfie feature
checkCameraAvailable method to check all browser, camera and permission settings before LIQA start
5.0.1, 5.0.2 - 2021-07-12Selfie capture bug on some older iOS versions
config.imageFormat is now stable
5.0.0 - 2021-06-26Web-worker for math parallel calculation is now loaded with all the rest files and initiated from global space
Canvas and videotag initialization are now available as an option outside of LIQA. The initialized elements now can be passed as parameters in liqa.init call. The default behaviour is still an automatic initialization of canvas and videotag inside of LIQA.
New feature - Autocapturing
LIQA can now take photos automatically if the image quality is good for the following 3 seconds.
Activate this feature by setting a new parameter autoSelfieCapture in config.json to true. LIQA will show a countdown overlay and perform selfie capturing by itself.
New feature - Landmarks (facial key points) prediction
LIQA can now return facial key points on a captured selfie image.
Activate this feature by setting a new parameter predictLandmarks in config.json to true. LIQA will return a set of facial key points - landmarks (see Selfie capturing results parsing for more details). See the Observables: Data Collection for more details about a new RX-JS observable with landmarks output.
Changed behaviour of captureSelfie function: now LIQA will always automatically stop videostream after capturing the selfie. To restart it simply call the liqa.play method.
Now the resolution of the selfie image returned by LIQA is limited by 2000 x 2000 pixels. This size is optimal for uploading to the server and suitable for any AI analysis.
Reworked qualityStatusAPI: from now the faceRotation field of the qualityStatus$ Observable contains 7 values with clear problem descriptions with head rotation. Now the usage of this for visual feedback mapping became much easier. For more see here Observables: Data Collection
We added new pages in our documentation:
If you are only thinking about trying LIQA, see here how to get it: Try LIQA: access request
If you are already developing your app with LIQA, see our updated Integration Guideline [JavaScript]
4.0.0 - 2021-05-18Improved stability by fixing issues with extreme memory consumption
Fixed issue with different FoV (field of view) on stream and final selfie image
Reworked API. Created method load, split methods init and play
Reworked input parameters and emitters' structures
Added new settings for LIQA illumination strictness
Moved demo page code inside of the package
3.1.0 - 2021-02-26Several more math elements were rewritten from JavaScript to WebAssembly scripts to speed up the processing. We observed a gain of 2-4 FPS on iPhone 8 and similar
Added check of technologies (WebAssembly, WebGL) availability in the browser during LIQA start
3.0.0 - 2021-02-19Fixed issue with incorrect timings of camera switches after captureSelfie action
Fixed problem with incorrect usage of a different camera for stream / selfie on devices with multiple real / virtual frontal cameras
Fixed problem with LIQA not running in Google Chrome on desktop
Updated FaceMesh dependencies to the latest version to fix raising warnings during package installation
Added watermark on stream for trial packages
Added an inner watchdog timer for LIQA restart in case of not responding
Modified Illumination estimation to better track different conditions
2.1.2 - 2021-01-19Fixed bug with incorrect work of config allowBeard / allowHairs parameters
Modified observables API to add one-line status
Added ability to turn on / off points visualization with config param showIlluminationPoints
New
From 5.4.0 LIQA always runs a full analysis on final image. The status of final image is returned to the same RxJS Observable qualityStatus$. You can read how to parse this observable here:
LIQA now supports back cameras! See the feature described in;