Preserving Product Fidelity in Large Scale Image Recontextualization with Diffusion Models
This addresses the challenge of creating diverse and realistic product visualizations for e-commerce and virtual showcasing, though it appears incremental by building on existing diffusion model techniques.
The paper tackles the problem of product image recontextualization by developing a framework that uses diffusion models and a novel data augmentation pipeline to generate high-fidelity images, showing effectiveness on datasets like ABO with realistic results for e-commerce applications.
We present a framework for high-fidelity product image recontextualization using text-to-image diffusion models and a novel data augmentation pipeline. This pipeline leverages image-to-video diffusion, in/outpainting & negatives to create synthetic training data, addressing limitations of real-world data collection for this task. Our method improves the quality and diversity of generated images by disentangling product representations and enhancing the model's understanding of product characteristics. Evaluation on the ABO dataset and a private product dataset, using automated metrics and human assessment, demonstrates the effectiveness of our framework in generating realistic and compelling product visualizations, with implications for applications such as e-commerce and virtual product showcasing.