📄 OralGPT  |  📄 OralGPT-Omni

🤗 MMOral-OPG-Bench  |  🤗 MMOral-Uni-Bench (Coming soon)

OralGPT: A Series Versatile Dental Multimodal Large Language Models

• OralGPT
• OralGPT-Omni
• OralGPT-Plus
• OralGPT-Patho
• OralGPT-3D
• OralGPT-Agent
• OralGPT-U
• ...

• News
• Overview
• Upcoming Updates
• Released Materials
• MMOral-Bench
• Citation

• [2025-12-17] 🔥🔥🔥 OralGPT-Omni-7B-Instruct has been released on 🤗 Hugging Face. 👏 Welcome to try it.
• [2025-11-27] 🔥🔥🔥 Our paper of OralGPT-Omni has been released on arXiv.
• [2025-10-31] 🔥 [NeurIPS 2025] MMOral-Bench (MMOral-OPG-Bench) has been released on 🤗 Hugging Face. 👏 Welcome to evaluate your LVLMs.
• [2025-09-22] 🚀 Multiple Dental Visual Expert Models have been released on 🤗 Hugging Face.
• [2025-09-19] 🎉 Our paper of OralGPT has been accepted by NeurIPS 2025.
• [2025-09-11] 🎉 Our paper of OralGPT has been released on arXiv.
• 🔜 We are actively developing MMOral-Bench v2, which will include:
  • ✅ More dental imaging modalities
  • ✅ Professional dentistry exam questions
  • ✅ Comprehensive evaluation of multiple MLLM performance in digital dentistry
• 🤝 For collaboration inquiries, please contact us at: 📮 isjinghao@gmail.com

OralGPT is a series multimodal large language model (MLLM) specialized in digital dentistry. It supports diverse dental imaging modalities, including:

• Intraoral images & videos
• Photographs
• Panoramic X-rays
• Periapical radiographs
• Cephalometric radiographs
• Histopathological images
• Textual Question & Conversation

OralGPT aims to be the foundation MLLM for AI-driven digital dentistry — bridging multimodal reasoning with clinical expertise. With Chain-of-Thought (CoT) reasoning, OralGPT simulates the diagnostic process of radiologists, ensuring outputs that are interpretable, trustworthy, and clinically reliable.

• 📦 Release of MMOral-Uni Benchmark
• 📑 Expanded instruction dataset with more diverse dental imaging modalities
• 🧪 Release of OralGPT-Plus

1. Multiple Dental Visual Expert Models released on 🤗 Hugging Face , covering detection, segmentation, and classification tasks in panoramic/periapical X-ray images.
2. [NeurIPS 2025] MMOral-Bench (MMOral-OPG-Bench) has been released on 🤗 Hugging Face.
3. 👉 Coming soon ...

Currently, you can evaluate your MLLM’s performance on panoramic X-ray analysis using MMOral-Bench.
All benchmark data are reviewed and validated by professional clinical dentists, ensuring accuracy and clinical reliability.

Our benchmark consists of both Open-Ended and Closed-Ended evaluation formats, with corresponding TSV files available at 🤗 Hugging Face.

For benchmark evaluation, we provide two approaches:

1. Using VLMEvalkit (supporting multiple pre-configured VLMs)
2. For VLMs not available in VLMEvalkit or new VLMs, we provide generic evaluation scripts: eval_MMOral_VQA_Closed.py and eval_MMOral_VQA_Open.py

Please refer to Evaluation Suite.

For models not supported by VLMEvalkit, you can use our generic evaluation templates. Simply add your model's inference method to either eval_MMOral_VQA_Closed.py or eval_MMOral_VQA_Open.py to conduct the evaluation. These scripts provide a flexible framework that can accommodate any VLM implementation.

#For Open-Ended Evaluation
python MMOral-Bench-EvalKit/eval_MMOral_VQA_Open.py \
  --benchmark_path '/path/to/your/MM-Oral-VQA-Open-Ended_processed.tsv' \
  --output_dir '/path/to/save/evaluation_results_open-4o' \
  --gpt_api_key 'your_api_key_here' \
  --gpt_api_base 'https://your-gpt-api-endpoint.com/v1/chat/completions' \
  --dataset_name 'MM-Oral-VQA-Open-Ended' \
  --model_name 'gpt4o'

#For Closed-Ended Evaluation
python MMOral-Bench-EvalKit/eval_MMOral_VQA_Closed.py \
  --benchmark_path '/path/to/your/MM-Oral-VQA-Closed-Ended.tsv' \
  --output_dir '/path/to/save/evaluation_results' \
  --api_url 'https://your-gpt-api-endpoint.com/v1/chat/completions' \
  --api_key 'your_api_key_here' \
  --dataset_name 'MM-Oral-VQA-Closed-Ended' \
  --model_name 'gpt4o'

This streamlined process allows you to easily benchmark any VLM against our MMOral-OPG-Bench.

If you find our work helpful, please cite us:

@article{hao2025mmoral,
  title={Towards Better Dental AI: A Multimodal Benchmark and Instruction Dataset for Panoramic X-ray Analysis},
  author={Hao, Jing and Fan, Yuxuan and Sun, Yanpeng and Guo, Kaixin and Lin, Lizhuo and Yang, Jinrong and Ai, Qi Yong H and Wong, Lun M and Tang, Hao and Hung, Kuo Feng},
  journal={arXiv preprint arXiv:2509.09254},
  year={2025}
}
@article{hao2025oralgpt,
  title={OralGPT-Omni: A Versatile Dental Multimodal Large Language Model},
  author={Hao, Jing and Liang, Yuci and Lin, Lizhuo and Fan, Yuxuan and Zhou, Wenkai and Guo, Kaixin and Ye, Zanting and Sun, Yanpeng and Zhang, Xinyu and Yang, Yanqi and others},
  journal={arXiv preprint arXiv:2511.22055},
  year={2025}
}
@article{hao2025oraldataset,
  title={Characteristics, licensing, and ethical considerations of openly accessible oral-maxillofacial imaging datasets: a systematic review},
  author={Hao, Jing and Nalley, Andrew and Yeung, Andy Wai Kan and Tanaka, Ray and Ai, Qi Yong H and Lam, Walter Yu Hang and Shan, Zhiyi and Leung, Yiu Yan and AlHadidi, Abeer and Bornstein, Michael M and others},
  journal={npj Digital Medicine},
  volume={8},
  number={1},
  pages={412},
  year={2025},
  publisher={Nature Publishing Group UK London}
}