LNL_K
June 22, 2023 ยท View on GitHub
Implementation of the paper "LNL+K: Learning with Noisy Labels and Noise Source Distribution Knowledge", [Arxiv]
Requirements
The code has been written using Python3 (3.10.4), run pip install -r requirements.txt to install relevant python packages.
Training
CIFAR dataset with synthesized noise.
Code has been modified from the original FINE implementation: https://github.com/Kthyeon/FINE_official/.
Please find the adaptation methods implementation in 'synthesized_data/trainer' folder.
- CRUST/CRUST+k: 'crust_trainer.py'
- FINE/FINE+K: 'dynamic_trainer.py'
- SFT/SFT+k: 'sft_trainer.py'.
Arguments settings & running experiments
Config files are in 'synthesized_data/hyperparams/multistep'.
Run bash scripts in 'synthesized_data/scripts/sample_selection_based'
Cell dataset BBBC036 with natural noise.
BBBC036 dataset is available at https://bbbc.broadinstitute.org/BBBC036.
Please find the adaptation methods implementation in 'adaptation_methods' folder.
- CRUST/CRUST+k: 'crust_k.py'
- FINE/FINE+K: 'dynamic_k.py'
- SFT/SFT+k: 'sft_k.py'.
Arguments settings & running experiments
Arguments are in 'cell_data/simple_multi_main.py'.
An example bash script is 'cell_data/test_main.sh'