Zero-To-CAD-100k

<p align="center"> <img src="assets/logo.png" alt="Zero-to-CAD" width="100%"/> </p> # Zero-to-CAD 100K **A curated subset of 100,000 geometrically diverse CAD construction sequences selected from Zero-to-CAD 1M.** <p align="center"> <img src="assets/agentic.png" alt="Zero-to-CAD agentic synthesis pipeline" width="800"/> </p> > **Zero-to-CAD: Agentic Synthesis of Interpretable CAD Programs at Million-Scale Without Real Data** > > [Mohammadmehdi Ataei](https://orcid.org/0000-0002-3399-9696), [Farzaneh Askari](https://orcid.org/0000-0003-0684-1102), [Kamal Rahimi Malekshan](https://orcid.org/0009-0004-1192-4724), [Pradeep Kumar Jayaraman](https://orcid.org/0000-0001-6314-6136) > > Autodesk Research ## Related Resources | Resource | Link | |----------|------| | 📄 **Paper** | [Zero-to-CAD: Agentic Synthesis of Interpretable CAD Programs at Million-Scale Without Real Data](https://arxiv.org/abs/2604.24479) | | 📦 **Zero-to-CAD 1M** (full dataset) | [ADSKAILab/Zero-To-CAD-1m](https://huggingface.co/datasets/ADSKAILab/Zero-To-CAD-1m) | | 📦 **Zero-to-CAD 100K** (this dataset) | You are here | | 🤖 **Fine-tuned Model** (Qwen3-VL-2B) | [ADSKAILab/Zero-To-CAD-Qwen3-VL-2B](https://huggingface.co/ADSKAILab/Zero-To-CAD-Qwen3-VL-2B) | | 🗂️ **Collection** | [ADSKAILab/Zero-To-CAD](https://huggingface.co/collections/ADSKAILab/zero-to-cad) | ## Overview This is the **curated 100K subset** of Zero-to-CAD, designed as an accessible entry point for researchers working with limited compute. The samples are selected for **maximum geometric diversity** from the full 1M dataset. ### Curation Process 1. **Visual embedding**: Each model is rendered from 8 viewpoints and encoded using DINOv3 features, averaged across views. 2. **Clustering**: K-means clustering partitions the embedding space into 100K clusters. 3. **Selection**: The nearest-to-centroid exemplar from each cluster is selected. This ensures the subset spans the full distribution of part types, operations, and geometric complexity present in the 1M dataset. <p align="center"> <img src="assets/samples.png" alt="Sample CAD models from Zero-to-CAD" width="100%"/> </p> ## When to Use This vs. the 1M Dataset | Use case | Recommended | |----------|-------------| | Quick prototyping & experimentation | ✅ **100K** | | Training large models | 📦 1M | | Benchmarking & evaluation | ✅ **100K** | | Resource-constrained environments | ✅ **100K** | | Maximum training data coverage | 📦 1M | ## Dataset Details ### Splits | Split | Samples | |-------|---------| | Train | 81,015 | | Validation | 9,734 | | Test | 9,767 | ### Data Fields Each sample contains: | Field | Type | Description | |-------|------|-------------| | `uuid` | `string` | Unique identifier (matches 1M dataset) | | `cadquery_file` | `string` | Executable CadQuery Python source code | | `num_faces` | `int` | Number of B-Rep faces in the final solid | | `face_latency_ms` | `float` | Time to compute face count (ms) | | `cadquery_ops_json` | `string` | JSON list of CAD operations used | | `cadquery_ops_count` | `int` | Number of CAD operations in the construction sequence | | `ops_latency_ms` | `float` | Time to extract operations (ms) | | `num_renders` | `int` | Number of rendered views | | `image_0` – `image_7` | `image` | 8 rendered views (256×256) | | `stl_file` | `bytes` | Exported STL mesh | | `step_file` | `bytes` | Exported STEP file | ### CAD Operations Coverage Broad operation vocabulary identical to the 1M dataset: - **Sketch primitives**: rect, circle, polygon, arc, spline, slot - **3D operations**: extrude, cut, revolve, loft, sweep - **Modifications**: fillet, chamfer, shell, offset - **Booleans**: union, cut, intersect - **Patterns**: linear, polar, mirror - **Features**: holes (through, blind, countersink), threads, ribs ## Quick Start ### Load the dataset ```python from datasets import load_dataset # Streaming mode — rows are fetched on demand ds = load_dataset("ADSKAILab/Zero-To-CAD-100k") # Get a single sample sample = next(iter(ds)) # Display the reconstructed script cad_code = bytes(sample["cadquery_file"]).decode("utf-8") print(cad_code) ``` ### Execute a sample ```python import cadquery as cq # Streaming mode — rows are fetched on demand ds = load_dataset("ADSKAILab/Zero-To-CAD-100k", split="train", streaming=True) sample = next(iter(ds)) # Execute the code from a sample code = bytes(sample["cadquery_file"]).decode("utf-8") exec(code) # Display generated CadQuery solid from IPython.display import display display(result) ``` ## Embeddings & FAISS Index Precomputed DINOv3 embeddings, a FAISS IVF-PQ index, and precomputed nearest neighbors for the full 1M dataset are available in the [1M dataset repository](https://huggingface.co/datasets/ADSKAILab/Zero-To-CAD-1m) under `embeddings/`. These cover all 1M samples, including every sample in this 100K subset. The `cad_gen_diverse_samples.csv` file in that folder documents the clustering-based selection process used to curate this subset (cluster IDs, distances to centroids, etc.). ```python import faiss from huggingface_hub import hf_hub_download # Download the FAISS index from the 1M repo index_path = hf_hub_download("ADSKAILab/Zero-To-CAD-1m", "embeddings/cad_gen_ivfpq.index", repo_type="dataset") index = faiss.read_index(index_path) ``` See the [1M dataset card](https://huggingface.co/datasets/ADSKAILab/Zero-To-CAD-1m#embeddings--faiss-index) for full usage examples. ## Intended Uses - **Quick prototyping** of CAD generation models with a manageable dataset size - **Evaluation & benchmarking** with a representative, diverse sample - **Fine-tuning smaller models** when full 1M training is not feasible ## Citation If you use this dataset, please cite: ```bibtex @misc{ataei2026zerotocadagenticsynthesisinterpretable, title={Zero-to-CAD: Agentic Synthesis of Interpretable CAD Programs at Million-Scale Without Real Data}, author={Mohammadmehdi Ataei and Farzaneh Askari and Kamal Rahimi Malekshan and Pradeep Kumar Jayaraman}, year={2026}, eprint={2604.24479}, archivePrefix={arXiv}, primaryClass={cs.CV}, url={https://arxiv.org/abs/2604.24479} } ``` ## License This dataset is released under the [Apache License 2.0](https://www.apache.org/licenses/LICENSE-2.0).