apple-aiml-research/ARKitScenes

This repo accompanies the research paper, ARKitScenes - A Diverse Real-World Dataset for 3D Indoor Scene Understanding Using Mobile RGB-D Data and contains the data, scripts to visualize and process assets, and training code described in our paper.

What it solves

It provides a massive, diverse real-world dataset of indoor scenes to improve 3D scene understanding, specifically addressing the need for high-quality RGB-D data captured from common mobile devices rather than just specialized equipment.

How it works

The project collects RGB-D data using the Apple LiDAR scanner on iPad Pros. It includes raw data, camera poses, surface reconstructions, and high-resolution depth maps from stationary laser scanners. For ground truth, it provides manually labeled 3D oriented bounding boxes for various furniture and room-defining objects.

Who it’s for

Researchers and developers working on 3D computer vision, indoor scene understanding, 3D object detection, and depth upsampling.

Highlights

  • The largest indoor 3D dataset of its kind, containing 5,047 captures of 1,661 unique scenes.
  • First RGB-D dataset captured with widely available mobile LiDAR scanners.
  • Includes high-resolution ground truth depth maps and manually labeled 3D bounding boxes.
  • Provides scripts and training code for 3D object detection and RGB-D guided upsampling tasks.

Related

  • Project
    ace-trump-tech/DeltaForce-OBS-Locker

    A real-time object detection and target locking system for Delta Force using YOLOv14 to identify game characters while filtering out environmental false positives.

  • Project
    foundry/OpenCVSwiftStitch

    An iOS example project that uses OpenCV to stitch images into panoramas, demonstrating modern Swift/C++ interoperability.

  • Project
    ros-perception/image_pipeline

    A ROS 2 package that processes raw camera images into a usable format for higher-level vision processing in robotics.

  • Project
    hcliucs/APSD

    A human-centric framework for assessing the stealthiness of adversarial perturbations in images, featuring a new dataset (APSD) and an attention-based prediction model (A2SM).