rishuKumar404/weasis-optimized-benchmark

Weasis Medical Imaging GUI Benchmark (Tabular Format)

Dataset Description

This dataset contains 267 end-to-end GUI automation tasks for the Weasis medical imaging viewer in tabular format, where each row represents one complete task with all associated data.

Dataset Summary

• Total Tasks: 267
• Total Images: 202
• Format: Tabular (each row = one task)
• Application: Weasis Medical Imaging Viewer
• Resolution: 1920x1080

Data Structure

Each row contains:

Column	Description	Type
serial_number	Task number (1-267)	int64
instruction	Natural language task description	string
json_task	Complete JSON data for the task	string
image_sequence	Screenshot sequence (→ separated)	string
images	All images for the task	List[Image]
task_id	Unique task identifier	string
num_steps	Number of steps in trajectory	int64
initial_image	Starting image filename	string
final_success	Whether task completed successfully	bool

Usage

from datasets import load_dataset
import json

# Load the dataset
dataset = load_dataset("rishuKumar404/weasis-tabular-benchmark")

# Access a task (row)
task_row = dataset["train"][0]
print(f"Task {task_row['serial_number']}: {task_row['instruction']}")
print(f"Steps: {task_row['num_steps']}")
print(f"Image sequence: {task_row['image_sequence']}")

# Parse the JSON task data
task_json = json.loads(task_row['json_task'])
print(f"Trajectory steps: {len(task_json['trajectory'])}")

# Access images
for i, image in enumerate(task_row['images']):
    if image is not None:
        print(f"Image {i+1}: {image.size}")

Task Examples

Row 1: Basic DICOM Loading

• Instruction: "Load CT abdomen series of Rishu, set a 1×2 layout, and invert contrast of one to compare them."
• Steps: 9
• Image sequence: "1.png → 2.png → Import DCM Slide CT Rishu.png → ..."
• Success: True

Row 25: Measurement Task

• Instruction: "Load chest X-ray of Rishu, use the Line tool to measure the heart width."
• Steps: 6
• Image sequence: "1.png → 2.png → ... → Line measurement.png"
• Success: True

Action Types

• CLICK: Button clicks, menu selections, dialog interactions
• SCROLL: Image navigation, panning, scrolling
• TEXT: Text input, annotations, search fields
• SEGMENT: ROI drawing, measurement tools, annotation drawing
• ZOOM: Zoom in/out operations
• COMPLETE: Task completion, saving, exporting

Advantages of Tabular Format

• Easy Analysis: Each task is one row
• Quick Filtering: Filter by instruction type, success rate, etc.
• Image Access: All images for a task in one place
• JSON Parsing: Full task data available when needed
• CSV Export: Can be opened in Excel/Google Sheets

Citation

@dataset{weasis_tabular_benchmark_2024,
  title={Weasis Medical Imaging GUI Benchmark (Tabular Format)},
  author={Rishu Kumar},
  year={2024},
  url={https://huggingface.co/datasets/rishuKumar404/weasis-tabular-benchmark}
}

License

MIT License