# LLAVABrain Dataset
---
license: mit
task_categories:
  - image-to-text
language_creators:
  - machine-generated
annotations_creators:
  - machine-generated
pretty_name: LLAVABrain
size_categories:
  - n<1K
---

# LLAVABrain Dataset

LLAVA-compatible brain tumor dataset...
This is a LLAVA-compatible brain tumor dataset in JSONL format with image-text pairs for multimodal training.

-  **Modality**: Image + Text (Conversations)
-  **Structure**:
  - `train.jsonl`: JSONL file where each entry contains:
    - `"image"`: path to image file (e.g., `images/1.jpg`)
    - `"conversations"`: list of message-response pairs
  - `images/`: contains 20 MRI image files
-    Compatible with LLaVA-style training.

Example:
```json
{
  "image": "images/1.jpg",
  "conversations": [
    {"from": "user", "value": "<image>\nDescribe tumor properties in JSON."},
    {"from": "assistant", "value": "{\n  \"tumor_type\": \"meningioma\",\n  \"tumor_size_cm\": 2.7,\n  \"location\": \"right cerebellopontine angle\",\n  \"edema\": false,\n  \"enhancement\": \"homogeneous\"\n}"}
  ]
}