HouseLayout3D is a benchmark dataset of architecturally complex real-world layouts.
With it, we introduce a training-free baseline method that outperforms state-of-the-art methods on both our new and existing datasets.
HouseLayout3D is a benchmark dataset of architecturally complex real-world layouts.
With it, we introduce a training-free baseline method that outperforms state-of-the-art methods on both our new and existing datasets.
Current 3D layout estimation models are primarily trained on synthetic datasets containing simple single room or single floor environments. As a consequence, they cannot natively handle large multi floor buildings and require scenes to be split into individual floors before processing, which removes global spatial context that is essential for reasoning about structures such as staircases that connect multiple levels. In this work, we introduce HouseLayout3D, a real world benchmark designed to support progress toward full building scale layout estimation, including multiple floors and architecturally intricate spaces. We also present MultiFloor3D, a simple training free baseline that leverages recent scene understanding methods and already outperforms existing 3D layout estimation models on both our benchmark and prior datasets, highlighting the need for further research in this direction.
We start from a set of RGB images and recover the key structural elements of the space:
walls, floors, ceilings, doors, windows, and stairs.
We introduce MultiFloor3D, a training-free baseline method that leverages recent advances in 3D reconstruction and 2D segmentation.
