Mohamed Benkedadra, a PhD student at UMONS, teamed up with researchers from UCL, ULB, and Multitel at the TRAIL Summer Workshop 2024 in Nantes, France. The workshop, renowned, gathering leading AI researchers from Belgian universities and research institutes. Mohamed and his team developed the CIA (Controllable Image Augmentation) framework using Stable Diffusion. This project, aimed at enhancing data augmentation, was later presented as a paper published in the MIPR 2024 conference. 🎉

The CIA framework addresses the limitations of traditional data augmentation methods by integrating Stable Diffusion models with ControlNet, generating synthetic images that are both diverse and realistic. This innovation allows researchers to enhance datasets for applications like object detection, image classification, and medical imaging, introducing new patterns and scenarios often missing from real-world data. 📊

During the workshop, Mohamed and his team focused on building a robust system that blends synthetic and real data to create powerful mixed datasets. The process involved using Stable Diffusion models conditioned on specific prompts and controlling the image creation with ControlNet, ensuring high fidelity and relevance for the targeted tasks. 🖼️

Mohamed presented the work after 2 weeks of coding and running experiments with the team. 🏆 This achievement underscores the value of collaborative research settings like the TRAIL Summer Workshop, where interdisciplinary teams develop groundbreaking solutions to complex challenges in AI and machine learning. 

As always, the workshop provided a lot of activities for researchers to network and foster a social environment to build long-lasting relationships and create interesting future collaborations !

Check out CIA on : https://github.com/multitel-ai/CIA/