L'Oréal evaluates their AI vision models with Giskard to improve customer experience

L'Oréal is one of the world's leading cosmetics companies, offering innovative products and services to meet the diverse beauty needs of consumers worldwide. As part of their initiatives to improve their customers’ digital experiences, L'Oréal sought to optimize the facial landmark detection models used in some of their applications like Skin Screen, Modiface, and Hapta.

Why L’Oréal and Giskard?

Facial landmark detection is a field in computer vision that identifies key points on a human face, enabling subsequent tasks like face reconstruction, identification, and emotion recognition. However, this process can be challenging due to variability in facial poses, lighting conditions, expressions, and other potential sources of bias.

To ensure reliable and inclusive predictions across all ages, genders, etc. L'Oréal needed a testing solution like Giskard. Their goal was to evaluate multiple open-source facial landmark detection models against a wide range of real-world scenarios and select the best-performing option for their digital services.

Evaluating Facial Landmark detection models with Giskard

L'Oréal worked with Giskard, leveraging our AI testing platform to conduct a granular comparison of facial landmark detection models. Giskard's API facilitated an in-depth evaluation across various criteria, including:

• Performance on partial facial regions
• Performance on face images with different head poses
• Robustness against image perturbations like blurring, resizing, and recolouring

Using Giskard's testing suite, L'Oréal could assess multiple models, including FaceAlignment, Mediapipe, and OpenCV, across diverse data slices and simulated real-world conditions.

Improving L’Oréal’s Customer experience and inclusivity

By leveraging Giskard's advanced testing capabilities, L'Oréal could make an informed decision on the most suitable facial landmark detection model for their applications. The comprehensive evaluation ensured that the selected model would deliver reliable and accurate predictions, improving the overall customer experience.

Moreover, Giskard's testing platform enabled L'Oréal to identify potential biases and address them proactively, ensuring consistent performance across diverse user demographics.

Integrating Giskard's automated evaluation reports into L'Oréal's development pipeline allowed for continuous monitoring and optimization of the facial landmark detection models. This process facilitated rapid iteration, enabling L'Oréal's team to make data-driven decisions quickly.

Giskard also provided L'Oréal with enterprise-grade assurances, ensuring that data stays within their environment. Our team closely collaborated with L'Oréal, offering dedicated support and guidance throughout the process.

What’s next?

L’Oréal’s partnership with Giskard exemplifies their commitment to delivering personalized and inclusive digital services to their customers.

As L’Oréal teams continuously strive to enhance the services offered to their customers, the integration of Giskard testing solution has become a crucial component. Combined with our expertise in AI quality, this enables L’Oréal’s AI teams to develop more accurate and reliable AI models.

“Through our collaboration with Giskard, we successfully developed and integrated a comprehensive package dedicated to keypoint detection model assurance. The package’s design enables us to assess the reliability of our models across various aspects in a straightforward, efficient, and user-friendly manner. Giskard provides a significant advantage in meeting the expectations of the EU AI Act.” - Alexandre Bouchez

Moving forward, L’Oréal will continue exploring innovative AI solutions and collaborating with Giskard to test and evaluate their models.

‍Reach out to us today to learn more about how we can help you to ensure your models are safe and reliable.