This directory contains a collection of Jupyter notebooks designed to guide you through various aspects of the framework -- from understanding the core concepts to visualizing results.
00_dataset.ipynb
Provides a basic introduction to the dataset, including how to load the data and visualize some example events.
Walks you visually through the tokenization strategy used in this work, starting with the usual tokenization strategy used in past works, and then moving on to the novel tokenization strategy (C-NMS) used in PoLAr-MAE.
Demonstrates how to pretrain a model in detail. Gives an overview of the architecture, as well as the training loop.
Demonstrates how to load pre-trained models and visualize the results of the pretraining process via (1) plotting group completions and (2) casting the embeddings to RGB space via PCA.
Demonstrates how to fine-tuning for semantic segmentation is done for this model in detail.
Demonstrates how to load fine-tuned models and visualize the segmentation results.
- Ensure all dependencies are installed as outlined in the main repository’s README.
- Activate your environment (e.g., using conda):
conda activate polarmae
- Launch Jupyter Notebook or JupyterLab within this directory:
jupyter notebook
- Open the notebook that interests you and follow the in-notebook instructions.
For further details on model configurations, experimental setups, and troubleshooting, please refer to the main documentation in the repository. If you have any questions or feedback, feel free to open an issue.
Happy learning!