InternScenes is a large-scale dataset comprising approximately 40,000 diverse indoor scenes and 1.96 million 3D objects. It covers 15 common scene types and 288 object classes, created by InternRobotics. The dataset was last updated on February 4, 2026.
Use Cases
- Train object detection models based on the 288 object classes.
- Generate realistic indoor environments for simulation based on the 40,000 diverse scenes.
- Benchmark scene understanding algorithms based on the dataset's complex layouts and preserved small items.
- Develop multimodal AI systems leveraging the dataset's 3D object and scene data.
Strengths
- Contains approximately 40,000 diverse indoor scenes.
- Includes 1.96 million 3D objects across 288 object classes.
- Described as roughly 10 times larger than existing datasets of this type.
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
- InternRobotics
- Freshness
- Last updated 2026-02-04 15:04:54; freshness should be verified.