img2cad-dataset: Dataset Card

Img2CAD Annotated CAD Dataset

This dataset contains annotated CAD models with semantic part information for training CAD reverse engineering models.

Dataset Description

The Img2CAD Annotated CAD Dataset is a comprehensive collection of 3D CAD models with: - Raw annotated CAD data in HDF5 format - Rendered images from multiple viewpoints - Semantic part labels and annotations - Train/test splits for three furniture categories

Categories

• Chair: 1,026 annotated chair models
• Table: 3,243 annotated table models
• Storage Furniture: 305 annotated cabinet/shelf models

4,574 models in total, as reported in the paper; train/test splits follow PartNet.

Data Structure

img2cad-dataset/
├── raw_annotated/           # Raw CAD data in h5 format
│   ├── chair/
│   │   ├── {id}/
│   │   │   ├── cad.h5       # CAD commands with part names
│   │   │   └── raw.obj      # Raw mesh (optional)
│   │   └── ...
│   ├── table/
│   └── storagefurniture/
├── blender_renderings/      # Rendered images (.png)
│   ├── {id}.png
│   └── ...
├── llamaft_gt_labels/       # Ground truth labels for LlamaFT
│   ├── chair/
│   │   ├── {id}.txt
│   │   └── ...
│   ├── table/
│   └── storagefurniture/
├── splits/                  # Train/test splits
│   ├── chair_train_ids.txt
│   ├── chair_test_ids.txt
│   └── ...
├── sym_labels/              # Symmetry labels (1=rotational, 2=reflection, 3=none)
├── trassembler_data/        # Preprocessed pickle files
│   ├── chair_pkl/
│   ├── table_pkl/
│   └── storagefurniture_pkl/
├── shapenet_partseg/        # Part segmentation annotations (unified txt format)
│   ├── synsetoffset2category.txt
│   ├── 03001627/            # Chair annotations
│   ├── 04379243/            # Table annotations
│   └── storagefurniture/    # Storagefurniture annotations
├── anno_id2model_id_*.json  # PartNet to ShapeNet ID mappings
└── *.json                   # Part name mappings and statistics

HDF5 File Format

Each cad.h5 file contains: - Part names as keys (e.g., "leg", "top", "back") - CAD command vectors following DeepCAD convention - Bounding box information ({part_name}_axis_aligned_bbox)

Usage

from huggingface_hub import hf_hub_download, snapshot_download

# Download specific files
cad_path = hf_hub_download(
    repo_id="qq456cvb/img2cad-dataset",
    filename="raw_annotated/chair/12345/cad.h5",
    repo_type="dataset"
)

# Or download entire directories
data_dir = snapshot_download(
    repo_id="qq456cvb/img2cad-dataset",
    repo_type="dataset",
    allow_patterns=["splits/*", "*.json"]
)

# Load with h5py
import h5py
h5 = h5py.File(cad_path, 'r')
part_names = [n for n in h5 if n != 'vec' and '_bbox' not in n]

Paper

Img2CAD: Reverse Engineering 3D CAD Models from Images through VLM-Assisted Conditional Factorization

• arXiv Paper
• Project Page

Citation

@inproceedings{you2025img2cad,
  title={Img2cad: Reverse engineering 3d cad models from images through vlm-assisted conditional factorization},
  author={You, Yang and Uy, Mikaela Angelina and Han, Jiaqi and Thomas, Rahul and Zhang, Haotong and Du, Yi and Chen, Hansheng and Engelmann, Francis and You, Suya and Guibas, Leonidas},
  booktitle={Proceedings of the SIGGRAPH Asia 2025 Conference Papers},
  pages={1--12},
  year={2025}
}

License

This dataset is released under the MIT License.

Acknowledgments

This dataset builds upon the PartNet and ShapeNet datasets. We thank the original authors for their contributions.