Project Hedgehog
This project involved a multi-stage pipeline focused on isolating individual subjects from a single source image containing a group of 10 people through precise bounding box annotation, ensuring each person was tightly and accurately localized for AI processing. Based on these 10 annotations, a generative AI model produced 40 composite images (4 per person), which were then meticulously curated down to a final set of 10 images by selecting the single best AI-generated background composite for each individual. Quality measures were strictly adhered to throughout, requiring bounding boxes to maintain tightness and full subject coverage to minimize background bleed, while the curation phase prioritized subject integrity, background realism, and aesthetic consistency to ensure natural and visually cohesive final outputs.