Our team trains and fine-tunes computer vision models for multiple use-cases with large and diverse datasets, ensuring that they learn to recognise and interpret visual information accurately. Continual learning, adaptation or bias mitigation are some of the challenges to be addressed.

To mitigate the lack of datasets with enough quality and volume for our customers’ use-cases, generation of synthetic information has become a fundamental objective for the team. We continuously explore and use the most advanced techniques in the state-of-the-art of Generative AI to produce high-quality, diverse, and representative datasets that can be used to train new models or to improve the accuracy of existing ones.

We complement our synthetic data generation capabilities with an internal Big Data infrastructure that is able to index a vast number of datasets including detailed information about the objects they contain thanks to the application of multi-modal vision and language models and auto-labelling techniques.

A strong emphasis is put on ethical considerations and trustworthiness. We are committed to ensuring that our computer vision solutions respect privacy, legal and ethical standards. We work also to apply the most recent MLOps advancements.