A Large-scale Aligned Shape Annotation on ARKitScenes is released!

apple / 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.

Other

635 stars 54 forks source link

A Large-scale Aligned Shape Annotation on ARKitScenes is released! #65

Open hugoycj opened 8 months ago

hugoycj commented 8 months ago

AI can now recreate detailed digital twins of your spaces! 🏠 We built LASA, 10K precise furnishing annotations aligned to most ARKitScenes scans. 🪑 Our DisCo method fuses inputs to reconstruct lifelike geometries. Plus improved detection builds complete home models!

Check https://x.com/ychngji6/status/1737341634004836690?s=20 for more Information. Your own digital home, immortalized! teaser

hugoycj commented 8 months ago

In addition to well-aligned CAD models handcrafted by artists, we introduce two novel methods, DisCo and OccGOD. DisCo enables high-fidelity 3D shape reconstruction from limited sensor data. OccGOD detects full objects from partial scene observations. Both techniques could advance 3D shape reconstruction and complete object detection capabilities for the platform like Vision Pro. By lifting reconstruction and detection quality to more competitive levels, DisCo and OccGOD promise more realistic and illuminating 3D spatial understanding for MR device users. result_1 result_2 results_3

littlejohntj commented 8 months ago

Does the Instance Reconstruction require RGB as part of the pipeline or can you you get the output based on just the Mesh alone?