Oral Cancer Classification — Hugging Face Model Repo
This repository contains the trained artifacts and supporting files for an oral cancer classification pipeline. The artifacts here are the result of a research/engineering pipeline that extracts features from medical images, selects important features, trains an attention-based classifier, and packages artifacts for inference and explainability.
This repo is intended to be stored as a private model repository on Hugging Face. The README documents the contents, how to run inference locally, how to upload the model to a private Hugging Face repo, and notes on reproducibility and licensing.
Repository contents
Files in this workspace (as of this snapshot):
feature_extract_SwinTransformer.pth— PyTorch checkpoint for the Swin Transformer based feature extractor used to create feature vectors from input images.SHAP_selected_features.npy- numpy array of features selected based on SHAP values during training.SHAP_500_selected_features.npy- numpy array of top 500 features selected based on SHAP values during training.vision_transformer.pt- PyTorch checkpoint for the Vision Transformer head used for classification.attention-cnn.pt- PyTorch checkpoint for the Attention CNN head used for classification.
If additional scripts or model weights are needed (for example the training scripts, dataset loaders, or the original image preprocessing code), they should be kept alongside this repo or referenced in a parent project — this repository stores model artifacts and supporting preprocessing objects required for inference and explainability.
Contact and citation
If you use this model or have questions please contract us.
Francis Rudra D Cruze - francisrudra@gmail.com.
If you use these models in academic research, please cite: