Evaluating and Optimizing GenAI Style-Transfer Pipelines for Industrial Use Cases

Flanders Make offers an internship focused on benchmarking and optimizing generative AI (GenAI) style-transfer and image-editing pipelines for industrial inspection systems. The project aims to improve the realism and variability of synthetic-to-real and real-to-real image translations to enhance training data quality for vision models in industrial applications.

• Industrial style-transfer challenge: The current pipeline uses diffusion models with masked guidance to transfer styles between synthetic and real images, producing annotated images valuable for training detection models, though improvements are possible.

• Internship goals: The internship involves benchmarking various GenAI methods, generating diverse masks, and reducing dependence on simulator data by increasing output robustness and variability.

• Learning outcomes: Participants will gain hands-on experience with diffusion models, fine-tuning style-transfer pipelines, handling domain shifts, creating randomized masks, using image-editing tools, and evaluating methods for industrial deployment.