README.md

---
configs:    
  - config_name: Example
    data_files: "example.parquet"
    description: "Combined multimodal example dataset (816 samples)."
    features:
      - name: input
        dtype: string
      - name: output
        dtype: string 
      - name: image
        dtype: image
      - name: type
        dtype: string
        
  - config_name: Science_MM
    data_files: "Science_MM/data.parquet"
    description: "Marine Science VQA (99 samples)."
    features:
      - name: id
        dtype: int64
      - name: pdf_title
        dtype: string
      - name: fig_name
        dtype: string
      - name: image
        dtype: image
      - name: question
        dtype: string
      - name: choices
        dtype: string
      - name: answer
        dtype: string
      - name: category
        dtype: float64
        



  - config_name: Science_Text
    data_files: "Science_Text/data.parquet"
    description: "Marine Science multiple-choice QA from academic PDFs (102 samples, 43 document sources)."
    features:
      - name: id
        dtype: int64
      - name: pdf_title
        dtype: string
      - name: question
        dtype: string
      - name: choices
        dtype: string
      - name: answer
        dtype: string
      - name: category
        dtype: string

license: mit
language:
  - zh
  - en
task_categories:
  - question-answering
  - image-text-to-text
pretty_name: OceanBenchmark
tags:
  - benchmark
  - evaluation
  - ocean
  - marine-science
---

# OceanBenchmark

## 1. Dataset Description

OceanBenchmark is a benchmark dataset designed to evaluate the comprehensive capabilities of marine-focused large models. It encompasses a diverse range of tasks, spanning from unimodal marine science knowledge question answering to complex multimodal visual question answering.

## 2. Sub-datasets

| Subset Directory | Task Type | Sample Size | Description |
|:---|---|---|---|
| **Example** | VQA | 816 | Combined multimodal examples with `Sonar` and `Bio` type labels. |
| **Science_Text** (`Ocean_Science_QA`) | QA | 102 | Text-only multiple-choice questions from marine science academic papers. |
| **Science_MM** (`Ocean_Science_VQA`) | VQA | 99 | Visual question answering based on scientific diagrams and imagery. |
| **Sonar** (`Sonar_VQA_Marine`) | VQA | 796 | Target detection and question answering evaluation on sonar imagery. |
| **Bio** (`Marine_Organisms_VQA`) | VQA | 472 | Classification and identification tests for marine organisms. |

## 3. Dataset Details

### Example
- **Sample count**: 816
- **Format**: Multimodal examples with input, output, image, and type fields

### Science_Text
- **Sample count**: 102
- **Source documents**: 43 unique PDFs
- **Format**: Multiple-choice QA (A/B/C/D)
- **Categories**: Physical Oceanography, Oceanic Climatology, Chemical Oceanography, Paleoceanography, Biological Oceanography (supports multi-label)
- **Note**: The `choices` field stores a dictionary string e.g., `{'A': 'option text', 'B': '...'}`. Use `ast.literal_eval()` to parse. Category annotations are available for 33 samples (32.4%).

### Science_MM
- **Sample count**: 99
- **Format**: Visual question answering with scientific figures
- **Features**: Includes image, question, choices, and answer fields

### Sonar
- **Sample count**: 796
- **Format**: Sonar image-based QA for target detection
- **Features**: input (question), output (answer), image

### Bio
- **Sample count**: 472
- **Format**: Marine organism classification VQA
- **Features**: input (question), output (answer), image

## 4. Usage Example

```python
from datasets import load_dataset
import ast

# Load the combined example subset
ds_example = load_dataset("zjunlp/OceanBenchmark", "Example", split="train")
print(ds_example[0]['input'])

# Load the text-only marine science QA subset
ds_qa = load_dataset("zjunlp/OceanBenchmark", "Science_Text", split="train")
sample = ds_qa[0]
print(f"Question: {sample['question']}")
print(f"Source: {sample['pdf_title']}")
choices = ast.literal_eval(sample['choices'])  # Parse the choices dictionary
for key, value in choices.items():
    print(f"{key}. {value}")
print(f"Answer: {sample['answer']}")
print(f"Category: {sample['category']}")
```

## Citation
If you use the data in your work, please cite:
```bibtex
@article{xue2026oceanpile,
  title={OceanPile: A Large-Scale Multimodal Ocean Corpus for Foundation Models},
  author={Xue, Yida and Zhang, Ningyu and Wu, Tingwei and Ma, Zhe and Ji, Daxiong and Wang, Zhao and Zheng, Guozhou and Chen, Huajun},
  journal={arXiv preprint arXiv:2605.00877},
  year={2026}
}
```