ArtClass v2: A Unified Framework for Generative Artistic Synthesis and Style Transfer via Latent Diffusion Models
In a physical art classroom setting, "Art Class 2" or "Paper Art" often refers to specific curriculum modules: artclass v2
Fine-grained visual classification (FGVC) of artwork is challenging due to high intra-class variance, subtle inter-class differences, and domain-specific attributes (e.g., brushwork, palette, era). We introduce , a new benchmark dataset consisting of 120,000 labeled artwork images spanning 150 artist styles, 12 historical periods, and 8 medium types (oil, watercolor, etc.). Unlike its predecessor, ArtClass v2 provides multi-label annotations (style + period + subject matter) and is designed to handle real-world art collection scenarios with class imbalance and partial labels. We evaluate 10 state-of-the-art FGVC architectures (e.g., DenseNet, Vision Transformers, MLP-Mixers) and show that even top models achieve only 68.3% top-1 accuracy, leaving significant room for improvement. ArtClass v2 is publicly available to spur research in computational art history and digital humanities. ArtClass v2: A Unified Framework for Generative Artistic
Early versions focused on basic link aggregation and simple site hosting. We evaluate 10 state-of-the-art FGVC architectures (e