[NEW RELEASE] visageISDK 9.1b3 introduces minimized jitter, better fitting and new lip landmarks

We recently conducted a survey to understand the specific areas of product improvement our customers were interested in. The responses affirmed our razor-sharp focus on areas such as reducing jitter, landmark point precision, and enhancing fitting quality.

What are the highlights of the new release?

With visageISDK 9.1b3, you can enjoy all the advantages of the stable 9.0 version, but with a more compact architecture that enhances face alignment performance. Moreover, this update brings significant progress in terms of accuracy and precision.

Let’s take a better look at what’s new:

➥ Significant jitter decrease

Switching to the new visage|SDK 9.1b3 release, you will enjoy improvement in face alignment. This is particularly useful if you require a non-smoothed facial points model and use the recently introduced temporal image denoising.

When compared to the 9.0 stable version, this release presents an all-inclusive package with increased stability in facial landmarks / significant reduction in jitter. If face alignment is crucial for your work, we highly recommend using this version.

To download the latest version, get in touch with us.

➥ Improved fitting quality

We have enhanced the fitting quality by simplifying the algorithm’s flow and expanding the range of Shape units calculation. These modifications significantly reduce fitting errors.

There is a minor trade-off of 2ms in performance, but it still maintains real-time performance in all use cases. To further boost performance, non-utilized Action and Shape units have been removed.

➥ Added lip landmark points

To enhance lip movement tracking and capture subtler expressions, we’ve added 16 new landmark points on the inner and outer lip line. This enhances the accuracy of lip tracking movement and improves the rendering of different masks. In total, there are now 115 facial points available for your applications.

Additional enhancements include:

  • Optimized HTML5 memory usage and working in strict mode. 
  • Fixed crashes in FaceDetector sample with multiple faces image

Try our newest edition for free

Why not try the latest and more efficient SDK for free? Simply get in touch to get started today.

For more interesting news and updates, follow us on LinkedIn, Twitter, Facebook, or Instagram, or simply subscribe to our newsletter.

Tags releases