Datasets:

SciVisAgentBench
/

SciVisAgentBench-tasks

This repository is a secondary repo of SciVisAgentBench, contains scientific visualization datasets and tasks for benchmarking scientific visualization agents.

Download Volume Datasets

download_and_organize.py downloads and organizes datasets under 512MB. Make sure you run the following script before you evaluate your agents through SciVisAgentBench:

python download_and_organize.py

Data Organization

All the volume datasets from http://klacansky.com/open-scivis-datasets/ have been organized into a consistent structure.

Directory Structure

The datasets and tasks for ParaView-MCP and ChatVis are organized into the main and sci_volume_data folders, while napari_mcp_evals holds tasks and datasets for napari-MCP.

Each dataset in the main and sci_volume_data folders follows this structure:

dataset_name/
├── data/
│   ├── dataset_file.raw         # The actual data file
│   └── dataset_name.txt         # Metadata about the dataset
├── GS/                          # Ground truth folder (ParaView state + pvpython code)
├── task_description.txt         # ParaView visualization task
└── visualization_goals.txt      # Evaluation criteria for the visualization

Available Volume Datasets

37 datasets under 512MB are suggested to be downloaded
18 datasets over 512MB are listed but not downloaded

See datasets_list.md for a complete list with specifications. And datasets_info.json is the complete JSON file with all dataset metadata.

Task Descriptions

Each dataset has:

Task descriptions - Based on dataset type (medical, simulation, molecular, etc.)
Visualization goals - Evaluation criteria tailored to the dataset characteristics
Ground Truth - Ground truth pvpython code, ParaView state and screenshots

Acknowledgement

SciVisAgentBench was mainly created by Kuangshi Ai (kai@nd.edu), Shusen Liu (liu42@llnl.gov), and Haichao Miao (miao1@llnl.gov). Some of the test cases are provided by Kaiyuan Tang (ktang2@nd.edu). We sincerely thank the open-source community for their invaluable contributions. This project is made possible thanks to the following outstanding projects: