ControlSketch

ControlSketch is a SketchDataset loader for the ControlSketch Dataset, introduced in the SwiftSketch paper (SIGGRAPH 2025).

Source: datasets/controlsketch.py

Data Format

The SketchKit loader downloads the original dataset zip file, extracts the cached SVG content, parses each SVG file into a list of SketchKit Path objects, and then wraps those paths into a Sketch instance. The SVG parsing step reads <path> elements using:

• M (Move)

• C (CubicBezier)

• L (Line, converted to an equivalent cubic Bézier segment)

Directory Layout

<root>/
  controlsketch_sketches/
    train/
      <category>/
        *.svg
    validation/
      <category>/
        *.svg
    test/
      <category>/
        *.svg

Splits are organized as train, validation, and test.

Code

from sketchkit.datasets import ControlSketch

ds = ControlSketch(cislab_source=True)

sketch = ds[0]
print(sketch.width, sketch.height)
print(sketch.path_num, sketch.curve_num)

bears = ds.items_metadata[
    (ds.items_metadata["category"] == "bear")
    & (ds.items_metadata["split"] == "validation")
]
bears_sketch = [ds[row.id] for _, row in bears[:100].iterrows()]

Arguments

• cislab_source: Selects the download source. If True, the dataset is downloaded from the CISLAB CDN mirror; otherwise it is downloaded from official host.

  • CISLAB mirror: https://cislab.hkust-gz.edu.cn/projects/sketchkit/datasets/ControlSketch/controlsketch_sketches.zip

  • Official host: https://drive.google.com/uc?id=1-kffChm6RXwTuP8qxUOT-QDf8-bSCH0w