Vitrus created a dataset pairing rendered tabletop scenes with CAD reference imagery for 3D pose grounding. Each training example includes a 32-view orientation bank for the queried part and exact geometric supervision from a manifest. The dataset was last updated on June 20, 2026.
Use Cases
- Train models for CAD-conditioned 3D pose grounding based on paired scene and CAD imagery.
- Develop render-and-compare orientation prediction systems using the 32-view orientation bank.
- Benchmark geometric reasoning algorithms with exact geometric supervision from manifests.
Strengths
- Provides exact geometric supervision from a manifest for precise training.
- Includes a structured 32-view orientation bank for each queried part.
Limitations
- Column-level documentation is absent; field semantics must be inferred after download.
- Row count is unknown, which may limit suitability assessment.
- Description metadata is limited; actual data quality requires manual inspection after download.
Provenance
- Source
- Vitrus, hosted on Hugging Face.
- Collection Method
- Synthetically generated, pairing rendered tabletop scenes with CAD reference imagery.
- Freshness
- Last updated 2026-06-20 12:17:34; freshness should be verified.