Datasets:

---

rajpurkarlab

/

3DReasonKnee

like 8

Follow

Rajpurkar Lab / Harvard Medical AI 96

This PR addresses several improvements for the 3DReasonKnee dataset card:

• Adds task_categories: image-text-to-text and object-detection are added to the metadata, aligning with the dataset's focus on grounded reasoning, diagnostic questions, and 3D bounding boxes in medical vision language models.
• Adds tags: The 3d-imaging tag is added to highlight the dataset's use of 3D MRI volumes for better discoverability.
• Links to the official Hugging Face paper: The paper 3DReasonKnee: Advancing Grounded Reasoning in Medical Vision Language Models is now linked directly in the introductory section, making it easier for users to access the research.
• Clarifies Content and Fixes Title: The card's main title typo ("Grouned" -> "Grounded") is corrected. The introduction is also rephrased to clearly state that this repository is the 3DReasonKnee dataset and describes its key components, while also noting the presence of code for reproduction and analysis.
• Adjusts "How to Use": This section is updated to emphasize that the processed dataset is readily available in /generate_data/, with separate instructions provided for those interested in reproducing it from raw OAI data.

These changes enhance the discoverability, clarity, and overall quality of the dataset card for researchers.