FedStyle: Style-Based Federated Learning Crowdsourcing Framework for Art Commissions
This addresses artists' concerns about privacy and competitiveness on art commission platforms, though it appears incremental as it adapts federated learning to a specific domain.
The paper tackles the problem of artistic style-based retrieval without exposing personal artworks by proposing FedStyle, a federated learning framework that addresses extreme data heterogeneity through abstract style representations and contrastive learning, demonstrating superiority in experiments.
The unique artistic style is crucial to artists' occupational competitiveness, yet prevailing Art Commission Platforms rarely support style-based retrieval. Meanwhile, the fast-growing generative AI techniques aggravate artists' concerns about releasing personal artworks to public platforms. To achieve artistic style-based retrieval without exposing personal artworks, we propose FedStyle, a style-based federated learning crowdsourcing framework. It allows artists to train local style models and share model parameters rather than artworks for collaboration. However, most artists possess a unique artistic style, resulting in severe model drift among them. FedStyle addresses such extreme data heterogeneity by having artists learn their abstract style representations and align with the server, rather than merely aggregating model parameters lacking semantics. Besides, we introduce contrastive learning to meticulously construct the style representation space, pulling artworks with similar styles closer and keeping different ones apart in the embedding space. Extensive experiments on the proposed datasets demonstrate the superiority of FedStyle.