Fine-grained visual categorization (FGVC) aims to recognize images belonging to multiple subordinate categories of a super-category (e.g. species of animals or plants, models of cars etc.) The difficulty lies with understanding fine-grained visual differences that sufficiently discriminate between objects that are highly similar in overall appearance but differ in fine-grained features. One of the key challenges in Fine-grained visual categorization is class imbalance—there is often rare classes for which only a few images are available, while the majority classes are overrepresented. This long-tailed distribution of datasets makes it difficult for models to learn adequately from underrepresented classes, leading to poor generalization and performance on rare categories. One promising approach to alleviate this imbalance is the use of generative models, such as Generative Adversarial Networks (GANs), Variational Autoencoders (VAEs), or diffusion models, to synthesize realistic images of rare classes. By generating synthetic data, we can balance these long-tailed datasets, ensuring that models are trained more equally across all classes. Conditional Image Generation offers a targeted approach by focusing on generating images that directly enhance the classifier’s performance on minority classes. This internship will explore cutting-edge solutions in generative AI for addressing class imbalance in FGVC. Core research questions include: How can GANs, VAEs, or diffusion models be adapted to generate high-quality images of rare classes in a fine-grained context? How do we ensure that generated images capture the subtle, fine-grained differences between classes? Can classification-guided image generation improve model performance specifically for underrepresented categories in long-tailed datasets? What impact does the use of synthetic data have on the overall classification accuracy and robustness? As an intern, you will play an integral role in the research and development of generative models to address this key problem. Responsibilities include: Conducting an in-depth literature review; Implementing and fine-tuning state-of-the-art generative models (GANs, VAEs, Diffusion models) for synthetic image generation; Benchmarking the performance of classification models trained with synthetic data against traditional methods; Documenting and presenting your results. Expected qualifications : strong interest in AI and Computer Vision, proficiency in Python and Pytorch, basic knowledge of Git for version control, a curious and autonomous mindset. Internship opportunity of 4 months duration. A CV and cover letter are requiered You will work in a dynamic and welcoming environment. You’ll be part of a friendly and energetic team. We offer a stimulating and supportive work environment. Recognized as a centre of excellence at the international level, Multitel develops and integrates emerging technologies in the industrial sector. These technologies are focused on five main areas of activity: Networks and Cybersecurity, Applied Photonics, Artificial Intelligence, Embedded Systems, and Railway Certification Within the Artificial Intelligence department, the AI team holds a crucial role in supporting companies with their technological innovation projects, guiding them from the feasibility phase to the minimum viable product in various application fields such as aerospace, automotive, Smart Cities, Industry 4.0, medical sciences, defense, and more. We love fresh ideas, shared coffees, and motivated people. If you’re eager to learn fast and make a real impact, join us!