Fusion 360 Gallery Dataset - build123d Reconstructions

7,683 parametric CAD models from the Fusion 360 Gallery Reconstruction Dataset reconstructed as executable build123d Python scripts.

Each model was originally a Fusion 360 sketch-and-extrude sequence. These scripts reproduce the geometry programmatically using build123d's BuildPart, BuildSketch, and BuildLine context managers, with NURBS curves handled via OCP bindings.

This is a Modified Set of the Fusion 360 Gallery Dataset. It is not the original dataset. See the original repository for the full dataset including assembly, segmentation, and reconstruction subsets with JSON metadata.

Structure

Models are bundled into batch files organized by complexity (timeline operation count):

01_2ops/       3,161 models   18 batches    2 timeline operations
02_3ops/         833 models    7 batches    3 timeline operations
03_4to5ops/    1,748 models   16 batches    4-5 timeline operations
04_6to7ops/      915 models   11 batches    6-7 timeline operations
05_8to10ops/     618 models   10 batches    8-10 timeline operations
06_11to15ops/    275 models    6 batches    11-15 timeline operations
07_16to20ops/     93 models    3 batches    16-20 timeline operations
08_21plus/        40 models    2 batches    21+ timeline operations

Each batch file contains up to ~200 models and stays under 5,000 lines. Every batch file is self-contained and executable.

Each batch also includes a contact sheet (batch_NNN_contact_sheet.png) showing 128x128 thumbnails of all models in a square grid:

Batch File Format

Each batch file contains:

1. Merged imports from all models in the batch
2. Model functions (def model_XXXXX():) with a # Description: comment and a docstring
3. MODELS dict mapping function names to expected volume and area
4. __main__ block that validates each model against expected measurements

Every model function has a visual description comment generated from a rendered screenshot:

# Description: A flat rectangular plate with rounded corners and four mounting holes near each corner.
def model_100221_4d7b66c4_0003():
    """Model: Gause Wrap"""
    with BuildPart() as part:
        with BuildSketch(Plane.XY):
            with BuildLine():
                ...
            make_face()
        extrude(amount=7.62)
    return part.part

MODELS = {
    "model_100221_4d7b66c4_0003": {"func": model_100221_4d7b66c4_0003, "volume": 0.3386, "area": 47.8954},
    ...
}

Usage

Install build123d (requires OCP/cadquery-ocp):

pip install build123d

Run a batch to validate all models:

python 01_2ops/batch_001.py

Import individual models:

from importlib import import_module

batch = import_module("01_2ops.batch_001")
shape = batch.model_100221_4d7b66c4_0003()
print(f"Volume: {shape.volume:.4f}")

Relationship to Source Dataset

• Source: Fusion 360 Gallery Reconstruction Dataset v1.0.1
• Scope: 7,683 of 8,625 reconstruction sequences (the ones that produce valid geometry)
• Transformation: Each JSON reconstruction sequence was converted to an equivalent build123d Python script. The JSON metadata files are not included here; refer to the original dataset for those.
• Naming: Function names encode the original file ID (e.g., model_100221_4d7b66c4_0003 corresponds to 100221_4d7b66c4_0003.json in the reconstruction dataset)
• Validation: Each model includes expected volume and surface area from the original Fusion 360 geometry, validated to within 1% tolerance

Citation

If this dataset contributes to your research, please cite the Fusion 360 Gallery Dataset:

@inproceedings{willis2021fusion,
  title={Fusion 360 Gallery: A Dataset and Environment for Programmatic CAD Construction from Human Design Sequences},
  author={Willis, Karl DD and Pu, Yewen and Luo, Jieliang and Chu, Hang and Du, Tao and Lambourne, Joseph G and Solar-Lezama, Armando and Matusik, Wojciech},
  booktitle={ACM SIGGRAPH},
  year={2021}
}

License

This Modified Set is distributed under the Fusion 360 Gallery Dataset License for non-commercial research purposes only. See LICENSE.md for full terms.