Chainization

Official pytorch implementation of ECCV 2022 paper, "Order Learning Using Partially Ordered Data via Chainization."

Dependencies

• Python 3.8
• Pytorch 1.7.1

Datasets

• MORPH II

• For MORPH II experiments, we follow the same fold settings in this OL repo.

• Adience

• For Adience experiments, we follow the official splits.

Quick Start : Train Model on Random Edge Cases

You can adjust supervision ratio by changing 'info_ratio' in the parse_option function.

• for Adience dataset

    $ python train_chainize_adience.py 

• for MORPH II dataset

    $ python train_chainize_morph.py

Referecences

1. FixMatch
2. POE