MedVLThinker-m23k-tokenized

Code: https://github.com/UCSC-VLAA/MedVLThinker
Project Page: https://ucsc-vlaa.github.io/MedVLThinker/

## 📊 Datasets

### Available Datasets

Our project provides several curated datasets for medical vision-language understanding and training:

| Dataset | Modality | Description | Download |
|---------|-----|-------------|----------|
| **MedVLThinker-m23k-tokenized** | Text-only | Tokenized version of the [m23k](https://github.com/UCSC-VLAA/m1) dataset | [🤗 HF](https://huggingface.co/datasets/UCSC-VLAA/MedVLThinker-m23k-tokenized) |
| **MedVLThinker-pmc_vqa-gpt_4o_reasoning-tokenized** | Image-Text | Tokenized PMC-VQA dataset with GPT-4o generated reasoning chains | [🤗 HF](https://huggingface.co/datasets/UCSC-VLAA/MedVLThinker-pmc_vqa-gpt_4o_reasoning-tokenized) |
| **MedVLThinker-pmc_vqa** | Image-Text |Processed PMC-VQA dataset for medical visual question answering with RLVR | [🤗 HF](https://huggingface.co/datasets/UCSC-VLAA/MedVLThinker-pmc_vqa) |
| **MedVLThinker-Eval** | Image-Text| Comprehensive evaluation dataset for medical VQA benchmarks | [🤗 HF](https://huggingface.co/datasets/UCSC-VLAA/MedVLThinker-Eval) |

### Dataset Usage

```python
from datasets import load_dataset

# Load evaluation dataset
eval_dataset = load_dataset("UCSC-VLAA/MedVLThinker-Eval")

# Load training dataset with reasoning
train_dataset = load_dataset("UCSC-VLAA/MedVLThinker-pmc_vqa-gpt_4o_reasoning-tokenized")

# Load PMC-VQA dataset
pmc_dataset = load_dataset("UCSC-VLAA/MedVLThinker-pmc_vqa")

# Load Medical23k tokenized dataset
m23k_dataset = load_dataset("UCSC-VLAA/MedVLThinker-m23k-tokenized")
```

<details><summary>Dataset details and preparation of your own</summary>

### Supported Datasets

Our framework supports evaluation on the following medical VQA datasets:

- **PMC-VQA**: PubMed Central Visual Question Answering
- **PathVQA**: Pathology Visual Question Answering
- **SLAKE**: Bilingual medical VQA dataset
- **VQA-RAD**: Radiology Visual Question Answering
- **MMMU Medical**: Medical subsets from MMMU benchmark
- **MedXpertQA**: Expert-level medical questions

### Data Format

All datasets follow a unified format:

```python
{
"images": [PIL.Image], # List of images
"question": str, # Question text
"options": Dict[str, str], # Multiple choice options
"answer_label": str, # Correct answer label (A, B, C, D)
"answer": str, # Full answer text
"reasoning": str, # Chain-of-thought reasoning (optional)
"dataset_name": str, # Source dataset name
"dataset_index": int # Unique sample identifier
}
```

代码仓库：https://github.com/UCSC-VLAA/MedVLThinker
项目主页：https://ucsc-vlaa.github.io/MedVLThinker/

## 📊 数据集

### 可用数据集

本项目提供多个经过精选的医学视觉语言理解与训练数据集：

| 数据集名称 | 模态 | 描述 | 下载地址 |
|---------|-----|-------------|----------|
| **MedVLThinker-m23k-tokenized** | 纯文本 | [m23k](https://github.com/UCSC-VLAA/m1) 数据集的分词版本 | [🤗 Hugging Face（HF）](https://huggingface.co/datasets/UCSC-VLAA/MedVLThinker-m23k-tokenized) |
| **MedVLThinker-pmc_vqa-gpt_4o_reasoning-tokenized** | 图像-文本 | 带有GPT-4o生成推理链的分词版PMC-VQA数据集 | [🤗 Hugging Face（HF）](https://huggingface.co/datasets/UCSC-VLAA/MedVLThinker-pmc_vqa-gpt_4o_reasoning-tokenized) |
| **MedVLThinker-pmc_vqa** | 图像-文本 | 基于RLVR处理的、用于医学视觉问答任务的PMC-VQA数据集 | [🤗 Hugging Face（HF）](https://huggingface.co/datasets/UCSC-VLAA/MedVLThinker-pmc_vqa) |
| **MedVLThinker-Eval** | 图像-文本 | 用于医学视觉问答基准测试的综合评估数据集 | [🤗 Hugging Face（HF）](https://huggingface.co/datasets/UCSC-VLAA/MedVLThinker-Eval) |

### 数据集使用方法

python
from datasets import load_dataset

# 加载评估数据集
eval_dataset = load_dataset("UCSC-VLAA/MedVLThinker-Eval")

# 加载带推理链的训练数据集
train_dataset = load_dataset("UCSC-VLAA/MedVLThinker-pmc_vqa-gpt_4o_reasoning-tokenized")

# 加载PMC-VQA数据集
pmc_dataset = load_dataset("UCSC-VLAA/MedVLThinker-pmc_vqa")

# 加载Medical23k分词数据集
m23k_dataset = load_dataset("UCSC-VLAA/MedVLThinker-m23k-tokenized")

<details><summary>数据集详情与自定义数据集制备</summary>

### 支持的数据集

本框架支持在以下医学视觉问答数据集上进行评估：
- **PMC-VQA**：PubMed Central 视觉问答数据集
- **PathVQA**：病理学视觉问答数据集
- **SLAKE**：双语医学视觉问答数据集
- **VQA-RAD**：放射学视觉问答数据集
- **MMMU Medical**：MMMU基准测试的医学子集
- **MedXpertQA**：专家级医学问答数据集

### 数据格式

所有数据集均遵循统一格式：
python
{
"images": [PIL.Image], # PIL.Image格式的图像列表
"question": str, # 问题文本（字符串类型）
"options": Dict[str, str], # 多项选择选项（字典格式，键为选项标签，值为选项文本）
"answer_label": str, # 正确答案标签（如A、B、C、D）
"answer": str, # 完整答案文本
"reasoning": str, # 思维链推理内容（可选字段）
"dataset_name": str, # 源数据集名称
"dataset_index": int # 样本唯一标识符（整数类型）
}

</details>