• Aggadi Gnanesh

Jewellery image generation requires synthesizing high-fidelity jewellery regions from celebrity images while preserving global facial structure and minimizing domain mismatch. Traditional GAN models struggle with maintaining spatial coherence and global consistency when generating high-resolution jewellery details. To address these limitations, we adopt COCO-GAN (COnditional COordinate GAN) and extend it with a cloudy-to-lucid multi-stage training strategy, enabling the model to progressively refine jewellery patches from low-resolution “cloudy” structures to high-detail “lucid” textures. Our enhanced COCO-GAN framework learns generation-by-parts using micro-patch-based synthesis, spatial coordinate conditioning, and macro-patch adversarial supervision. This approach allows scalable training, reduced memory cost, beyond-boundary generation, and accurate jewellery reconstruction even at resolutions exceeding the training distribution. Our experiments on the Celeb-Jewellery dataset demonstrate that the proposed system generates visually coherent, globally consistent, and high-resolution jewellery images at 256×256 and beyond.

GAN, COCO-GAN, Coordinates, Patch-based Generation, Image Synthesis, Jewellery Reconstruction, Beyond-Boundary Generation

"JewelGAN: A Coordinate-Aware Generative Model for High-Resolution Jewelry Image Synthesis", International Journal of Emerging Technologies and Innovative Research (www.jetir.org), ISSN:2349-5162, Vol.12, Issue 12, page no.c612-c621, December-2025, Available :http://www.jetir.org/papers/JETIR2512269.pdf

"JewelGAN: A Coordinate-Aware Generative Model for High-Resolution Jewelry Image Synthesis", International Journal of Emerging Technologies and Innovative Research (www.jetir.org | UGC and issn Approved), ISSN:2349-5162, Vol.12, Issue 12, page no. ppc612-c621, December-2025, Available at : http://www.jetir.org/papers/JETIR2512269.pdf