Img2Cadseq: Synthetic and Real-World CAD Data for Image-to-CAD Generation

Img2CADSeq is a dataset introduced for a SIGGRAPH 2026 paper to train and evaluate an image-to-CAD generation pipeline. It includes two distinct data types: curated synthetic models and real-world captured objects. The synthetic component, CAD-220K, is a curated subset of the ABC dataset.

Use Cases

• Training image-to-CAD generation models based on the described synthetic and real-world data.
• Evaluating point cloud lifting stages in 3D reconstruction pipelines based on the CAD-220K subset.
• Benchmarking sequence-based diffusion models for standardized STEP file output.

Strengths

• Includes two distinct data types: curated synthetic models and real-world captured objects.
• CAD-220K subset is derived from the established ABC dataset.
• Designed to support a SIGGRAPH 2026 publication, suggesting a research-grade purpose.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count, file formats, and license information are unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

Author Rilpraatsy on Hugging Face, associated with the official repository for the Img2CADSeq project.

Collection Method

Likely contains a mix of programmatically generated synthetic data and captured real-world objects.

Freshness

Last updated 2026-06-10 08:07:27; freshness should be verified.