SLAKE: A Semantically-Labeled Knowledge-Enhanced Dataset for Medical Visual Question Answering

The report of SLAKE: A Semantically-Labeled Knowledge-Enhanced Dataset for Medical Visual Question Answering [ISBI 2021 oral]. We build a large bilingual dataset, SLAKE, with comprehensive semantic labels annotated by experienced physicians and a new structural medical knowledge base for Med-VQA. Besides, SLAKE includes richer modalities and covers more human body parts than the currently available dataset

>Download

[News] SLAKE 1.0 can be downloaded from here (or Google Drive, Huggingface).
(We have cleaned up the dataset again, which may be slightly different from the original paper. Any questions or mistakes, please contact us. Thank you!!! )

[News] Slake 2.0 is in preparation, please look forward to it!

>Features

• Visual Annotations
  1. Masks for semantic segmentation
  2. Bounding boxes for object dection
• Diverse Questions
  1. Knowledge-based Questions: involve external medical knowledge (solved by provided medical knowledge graph)
  2. Bilingual : English & Chinese
• Large Scale
  1. More question-image pairs
  2. More body parts: brain, neck, chest, abdomen, and pelvic cavity

>Contributors

• Bo Liu
• Li-Ming Zhan
• Li Xu
• Xiao-Ming Wu (Supervisor)