Dataset

SceneFun3D

Contains 710 indoor scene groups with Faro point clouds and several RGB-D iPad videos per scene, annotated for fine-grained interactive parts and tasks.

Hugging Face ↗Jun 29, 2026Code ↗

  • cv
  • robotics
  • multimodal
  • affordance
  • point-cloud
  • rgbd-video

What it is

3D indoor scenes with fine-grained interactive parts, affordances, motion, and language, paired with RGB-D video views and 2D/3D projections. 710 indoor scenes with high-res Faro point clouds plus RGB-D iPad videos; 14,867 annotated interactive elements with 9 affordance classes, motion parameters, and task descriptions, linked across 3D boxes and 2D frame projections.

Key results

  • 710 real-world indoor scenes
  • 14,867 functional element annotations, 9 affordance classes
  • RGB-D videos + 3D point clouds; CC BY-NC-SA 4.0

Useful for 3D affordance grounding, manipulation target localization, and motion prediction in indoor scenes.

One useful ML email a week.

Worth your time, not all of your time.

We track new ML work across papers, code, and data, then keep only the pieces worth a closer look. Selected, not collected.

Papers.Data.Code

The Weekly ML Signal

Selected, not collected.

One useful ML email a week: the papers, repos and datasets worth your attention.

Worth your time, not all of your time.