cchar

This project is for Learning Part-whole Hierarchies from the Sequence of Handwriting.

• Data

  • Please download CChar dataset from google drive.
  • It includes
    • $images$: 73,086 images for training visual feature extractors;
    • $annotations$: annotations for image classification and sequence generation.
    • $feats$: VGG/MAE/ViT visual feature files for sequence generation.

• Experiment:

  • Visual feature extraction:
    • MAE(ViT-based) models and visual features are based on official code.
  • Sequence generation:
    • Place different feature zip files under ./cchar_seq/data/feats and unzip them. For example, /cchar_seq/data/feats/tinyvgg_256.
    • Place different json files (random splits for train/val/test) under ./cchar_seq/data for experiment 3.

• Versions:

  • python>=3.7
  • pytorch=1.13.1