Grounded Teacher (GT)

This repository contains model weights and resources for Grounded Teacher, a framework designed to tackle Source Free Object Detection (SFOD), especially under domain shift and context imbalance conditions—common challenges in medical imaging.

🧠 Abstract

We focus on the Source Free Object Detection (SFOD) problem, where source data is unavailable during adaptation, and the model must adapt to an unlabeled target domain. In medical imaging, many approaches rely on semi-supervised student-teacher setups to handle domain gaps.

However, context imbalance and domain shifts often lead to biased teacher models, producing poor pseudolabels and causing the student model to collapse.

To address this, we introduce Grounded Teacher (GT):

Models contextual relationships via a relational context module.
Applies targeted augmentations for underrepresented classes across/within domains.
Adds an expert foundational branch to supervise the student model and enhance prediction quality.

GT shows significant improvement across three medical imaging datasets, with thorough ablation studies validating its effectiveness.

📦 Resources

🔗 GitHub: Grounded Teacher on GitHub
📄 Paper: Context Aware Grounded Teacher for Source Free Object Detection

👨‍🔬 Authors

Tajamul Ashraf
Rajes Manna
Partha Sarathi Purkayastha
Tavaheed Tariq
Janibul Bashir

📅 Publication

Title: Context Aware Grounded Teacher for Source Free Object Detection
Published: April 22

📃 Citation

@misc{ashraf2025contextawaregroundedteacher,
      title={Context Aware Grounded Teacher for Source Free Object Detection}, 
      author={Tajamul Ashraf and Rajes Manna and Partha Sarathi Purkayastha and Tavaheed Tariq and Janibul Bashir},
      year={2025},
      eprint={2504.15404},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2504.15404}, 
}