Skin SaaS algorithms updates

Quality algorithm update and Skintone Classification release

Quality algorithm update

Attention to API users: Quality Algorithm update is coming in 30 days.

What will change? The update is focused on making the output more clear and detailed. The update affects submetrics and feedback outputs of the algorithm. Read more below ⬇️

No actions are required if you:

  • Only use a B2C app constructor;

  • Only use the main metric output.

Upcoming changes

Submetrics

BeforeAfter

Previously, submetrics were included as key-value pairs (name: value) in a separate field QualityResult.sub_metrics.

Now, the submetrics’ output has the same structure as the other algorithms’ output in Face Metrics 2.0.

They are now included as a list of Metrics (with value, tech_name and widget_type fields) in QualityResult.area_results[0].sub_metrics field

Submetrics were also redesigned to better reflect image quality properties for the end-user:

  1. The “detection” submetric was removed;

  2. To replace the “detection” submetric, two new submetrics were introduced: “full_face_score” - the degree of face presence and “occlusion_score” - the degree of skin areas occlusion.

  3. A new submetric “shadows_score” is added that indicates the degree of the skin illumination uniformness and the presence of shadows.

Feedback

BeforeAfter

Previously, feedback was included as key-value pairs (name: value) in a separate field QualityResult.feedback.

Feedback -> Problems* “Problems” output is designed to better reflect image quality issues for the end-user. Problems are presented as a list of short codename strings in a new field QualityResult.problems. *The old field named “feedback” is deprecated and will be removed.

All details regarding the update can be found in our documentation.

Please contact us via the support desk if you have any further questions.

Skin Tone Classification

The new skintone classification algorithm has a more clear output than the old Skintone algorithm.

Skintione Classification algorithm analyses how tan the skin is and returns the result as one of the 6 values:

  1. Very light - very light skin tone

  2. Light- light skin tone

  3. Intermediate - intermediate skin tone

  4. Tan - tan skin tone

  5. Brown- brown skin tone

  6. Dark - dark skin tone

Read more about the new algorithm in our documentation.

Other improvements

✨ We are constantly improving our algorithms and in 30 days we are going to release a new version of the perceived age algorithm!

It is trained on more data and gives more precise results!

Last updated