Image Aesthetic Assessment in PyTorch

Image Aesthetic Assessment in PyTorch with implemented popular datasets and models (possibly providing the pretrained ones).

Note that this is a work in process.

Supported Datasets

• AVA Datasets A large-scale database for aesthetic visual analysis
  • 250,000 images, 235556 (training) and 19926 (testing)
  • aesthetic scores for each image
  • semantic labels for over 60 categories
  • labels related to photographic style
• AADB Datasets A aesthetics and attributes database
  • 10,000 images in total, 8458 (training) and 1,000 (testing)
  • aesthetic scores for each image
  • meaningful attributes assigned to each image (TODO)
• CHAED A chinese handwriting aesthetic evaluation database
  • 1000 Chinese handwriting images
  • diverse aesthetic qualities for each image from three level
  • rated by 33 subjects

Supported Models

• Unified Net Image Aesthetic Assessment Based on Pairwise Comparison – A Unified Approach to Score Regression, Binary Classification, and Personalization (In Progress)
• PAC-Net PAC-NET: PAIRWISE AESTHETIC COMPARISON NETWORK FOR IMAGE AESTHETIC ASSESSMENT (TODO)
• NIMA NIMA: Neural Image Assessment
• MP_ada MP_ada: Attention-based Multi-patch Aggregation for Image Aesthetic Assessment

How to use

python -u train.py --config path/to/your/config