Valeo uses brighter AI’s Deep Natural Anonymization for first extensive automotive fisheye dataset

The Client

Valeo is a global automotive supplier headquartered in France with €19 billion in annual revenues. The company provides a wide range of products to automakers and the aftermarket in 33 countries with 100,000+ employees and 187 production sites. With 63 R&D centers worldwide, Valeo is investing significantly into the future of mobility and with its leadership in sensors, it is working at the forefront of autonomous driving.

Privacy by design to enable data sharing

Compatible with fisheye image data

Vehicle data from various countries

The Challenge

For autonomous driving research and training of neural networks and validation systems, large amounts of image data are needed. Therefore, Valeo has created the WoodScape dataset. What makes the dataset particularly valuable is that it is the first extensive automotive fisheye dataset, consisting of images from four surround-view cameras that have been collected in several countries. Due to strict privacy regulations in the respective countries and Valeo's focus on privacy compliance and social responsibility, image anonymization was an important priority. At the same time, "traditional approaches like pixelating causes artifacts in the image and can have a significant negative impact on the quality of the trained model" (Valeo, 2021). Thus, the challenge was to anonymize the data while delivering the unique value of the WoodScape dataset, including the semantic segmentation annotation and ML and analytics compatibility. Therefore, a natural appearance and minimal pixel impact on the visual data is important.

The Solution

With Deep Natural Anonymization, the advanced anonymization technique of brighter Redact, we were able to fulfill the complex needs of Valeo for the dataset. Because the solution is camera-agnostic, it works for any kind of setting and format and thus has no problem with the fisheye format. Personally identifiable information is detected accurately and redacted with a generated replacement. This enables annotation, analytics, and machine learning use-cases while being privacy-compliant. The flexible deployment options of brighter Redact give the freedom to either anonymize in the cloud on certified servers or on-premise where the data resides. Valeo chose the latter for full control over the environment.

Read about the compatibility of Deep Natural Anonymization with machine learning.

Learn about our extensive experience in the automotive sector.




The Result

After a successful proof of concept with initial data, Valeo and brighter AI agreed on a longer-term engagement in order to handle current and future data privacy requirements for visual data. This made the creation of the WoodScape dataset possible, which now consists of over 10,000 images with semantic annotation of 40+ classes at the instance level. Once again, it was proved that brighter Redact with its unique Deep Natural Anonymization is the ideal redaction software for automotive training data, providing the highest accuracy without any negative impact on machine learning models.

Find the dataset on GitHub.

Learn more about brighter Redact.

"WoodScape has publicly collected image data from several countries and there is a significant risk of violating privacy regulations. Anonymizing personally identifiable information like faces and license plates with traditional approaches like pixelating causes artifacts in the image and can have a significant negative impact on the quality of the trained model. To tackle the dilemma, we made use of brighter AI’s Deep Natural Anonymization."