RGB Kitchen Video Dataset
RGB video from real kitchen environments with monocular depth estimates provided as an annotation layer for training depth-aware kitchen robots. Capture is RGB video — Claru does not capture hardware depth.
Dataset at a Glance
Comparison with Public Datasets
How Claru's dataset compares to publicly available alternatives.
| Dataset | Clips | Hours | Modalities | Environments | Annotations |
|---|---|---|---|---|---|
| ScanNet | 1.5K scans | ~5 | RGB-D | 707 rooms | 3D segmentation |
| NYU Depth V2 | 1.4K | ~2 | RGB-D | 464 scenes | Depth, segmentation |
| Claru RGB Kitchen Video | Scoped per project | Scoped per project | RGB (depth as derived layer) | Real kitchens | Monocular depth estimates, objects, surfaces |
Use Cases
Depth-Aware Manipulation
Using derived depth estimates for reach planning and collision avoidance in cluttered kitchens. Example models: UniDepth, DepthAnything, ICP grasping.
3D Scene Understanding
Building 3D kitchen representations for spatial reasoning. Example models: ScanNet++, Habitat, iGibson.
Transparent Object Detection
Labelling glass and clear objects from RGB appearance, where depth estimates are least reliable. Example models: ClearGrasp, TransCG, Dex-NeRF.
Key References
- [1]Dai et al.. “ScanNet: Richly-annotated 3D Reconstructions of Indoor Scenes.” CVPR 2017, 2017. Link
- [2]Yang et al.. “Depth Anything: Unleashing the Power of Large-Scale Unlabeled Data.” CVPR 2024, 2024. Link
- [3]Sajjan et al.. “ClearGrasp: 3D Shape Estimation of Transparent Objects.” ICRA 2020, 2020. Link
How Claru Delivers This Data
Capture is RGB video. Monocular depth estimates may be provided as an annotation layer. Contact Claru to scope RGB capture and derived annotation requirements.
Frequently Asked Questions
No. Capture is RGB video — Claru does not deploy depth sensors. Monocular depth estimates computed from the RGB frames may be provided as an annotation layer; they are relative estimates, not registered sensor depth.
Monocular depth estimates are least reliable on glass, mirrors, and polished steel. Where a depth layer is scoped, those regions can be flagged as low-confidence, and transparent objects can receive boundary annotations.
Only approximately. Monocular depth estimates are relative, not metric, so point clouds back-projected from them carry scale ambiguity. Teams that need metric 3D reconstruction should treat the RGB video as input to their own reconstruction pipeline rather than as sensor ground truth.
Request a Sample Pack
Get a curated sample of RGB kitchen video with derived depth and annotations to evaluate for your project.