SSL Study

Work in progress...

Dataset

INFO about the dataset.

We have logged the entire dataset as W&B Artifacts for building easy data pipeline for our study. This also enabled us to download the dataset on any machine easily. Here's the Kaggle kernel used to log them as W&B Artifacts: Save the Dataset as W&B Artifacts.

[Add chart of the dataset with associated W&B Tables view]

Usage

Installations

• Clone the repo: git clone https://github.com/cosmo3769/SSL-study
• Move into the repo: cd SSL-study
• Run: python setup.py install. If you want to develop do: pip install -e .
• Run: pip install --upgrade -r requirements.txt

Wandb Authorization

• Run: bash ssl_study/utils/utils.sh

Supervised Pipeline

To train the supervised pipeline that trains a baseline image classifier using labeled training dataset:

python train.py --config configs/baseline.py

• --wandb: Use this flag to log the metrics to Weights and Biases
• --log_model: Use this flag for model checkpointing. The checkpoints are logged to W&B Artifacts.
• --log_eval: Use this flag to log model prediction using W&B Tables.

To test your trained model, run:

python test.py --config configs/test_config.py

Sweeps

• Run: python sweep_train.py --config configs/baseline.py
• Run: wandb sweep /configs/sweep_config.yaml
• Run: wandb agent entity-name/project-name/sweep-id

NOTE

• Change the entity-name, project-name, and sweep-id according to your entity-name, project-name, and sweep-id.
• You will get your sweep-id by running wandb sweep /configs/sweep_config.yaml as mentioned above.

Tests

To run a particular test: python -m unittest tests/test_*.py

SimCLRv1 pretrain

Run: python simclrv1_pretext.py --config configs/simclrv1_pretext_config.py --wandb --log_model

SimCLRv1 train

Run: python simclrv1_downstream.py --config configs/simclrv1_downstream_config.py --model_artifact_path <path/to/model/artifact>

Citations

@misc{su2021semisupervised,
      title={The Semi-Supervised iNaturalist-Aves Challenge at FGVC7 Workshop}, 
      author={Jong-Chyi Su and Subhransu Maji},
      year={2021},
      eprint={2103.06937},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}