michaelrodcs/art-style-convnext

🎨 ArtEra: ConvNeXt-based Art Style Classifier

Model Description

ArtEra is a computer vision model fine-tuned to classify 21 different artistic styles. It is based on the modern ConvNeXt-Tiny architecture (pre-trained on ImageNet-1K) and optimized using a progressive resolution training strategy.

Intended Uses & Limitations

Intended Use

• Art History Tools: Automating the tagging of large collections of paintings.
• Academic Research: Analyzing stylistic trends and connections between movements.

Limitations and Bias

• Style Boundaries: The model may struggle with transitional periods (e.g., late Realism vs. early Impressionism).
• Dataset Bias: Bias towards Western art history due to the nature of the WikiArt collection.

Training Data

The model was trained on a curated subset of the WikiArt dataset, filtered into 21 balanced classes (approx. 76,000 images).

Included Styles: Abstract Expressionism, Art Nouveau Modern, Baroque, Color Field Painting, Cubism, Early Renaissance, Expressionism, Fauvism, High Renaissance, Impressionism, Mannerism Late Renaissance, Minimalism, Naive Art Primitivism, Northern Renaissance, Pop Art, Post Impressionism, Realism, Rococo, Romanticism, Symbolism, Ukiyo-e.

Evaluation Results

• Top-1 Validation Accuracy: 68.76%
• Best Performing Styles: Cubism (91%) and Pop Art (90%).

How to Use

This model is designed to be loaded using torchvision. You must modify the final classification layer of a convnext_tiny model to match the 21 output classes before loading the weights.

@misc{artera-convnext,
  author = {Michaelrodcs},
  title = {ArtEra: A ConvNeXt-based Art Style Classifier},
  year = {2026},
  publisher = {Hugging Face},
  howpublished = {\url{[https://huggingface.co/michaelrodcs/art-style-convnext](https://huggingface.co/michaelrodcs/art-style-convnext)}}
}