Infinigen Indoors: Photorealistic Indoor Scenes using Procedural Generation

Infinigen Indoors: Photorealistic Indoor Scenes using Procedural Generation

17 Jun 2024 | Alexander Raistrick*, Lingjie Mei*, Karhan Kayan*, David Yan, Yiming Zuo, Beining Han, Hongyu Wen, Meenal Parakh, Stamatios Alexandropoulos, Lahav Lipson, Zeyu Ma, Jia Deng
Infinigen Indoors is a Blender-based procedural generator of photorealistic indoor scenes. It expands the Infinigen system, which focuses on natural scenes, to indoor scenes by introducing a diverse library of procedural indoor assets, including furniture, architecture elements, appliances, and other day-to-day objects. It also introduces a constraint-based arrangement system, which allows users to specify scene composition constraints through a domain-specific language and a solver that generates scene compositions that maximally satisfy the constraints. Infinigen Indoors is open-sourced under the BSD license and allows generated 3D objects and scenes to be directly used for training embodied agents in real-time simulators such as Omniverse and Unreal. The system provides a constraint-based arrangement system that enables users to specify scene arrangement constraints, including symmetry, spatial relations, quantity, physics, and accessibility. The system also includes a constraint solver that searches for arrangements that maximally satisfy the constraints. Infinigen Indoors generates diverse indoor objects and full indoor scenes with physically and semantically plausible arrangements. It is fully procedural, using no external assets and only mathematical rules to generate everything from scratch. The system includes a library of procedural generators of indoor assets, a constraint-based arrangement system, and a tool to export generated scenes to real-time simulators. The system is designed to be highly customizable and allows users to specify high-level goals for abstract classes of objects. The system is also capable of generating realistic indoor scenes with diverse room combinations, connectedness, and overall contours. The system has been validated through experiments on tasks such as shadow removal and occlusion boundary estimation, demonstrating its effectiveness in generating synthetic data for training embodied agents. Infinigen Indoors is open-sourced under the BSD license to enable free and unlimited use by everyone and to enable community contributions of additional procedural generators.Infinigen Indoors is a Blender-based procedural generator of photorealistic indoor scenes. It expands the Infinigen system, which focuses on natural scenes, to indoor scenes by introducing a diverse library of procedural indoor assets, including furniture, architecture elements, appliances, and other day-to-day objects. It also introduces a constraint-based arrangement system, which allows users to specify scene composition constraints through a domain-specific language and a solver that generates scene compositions that maximally satisfy the constraints. Infinigen Indoors is open-sourced under the BSD license and allows generated 3D objects and scenes to be directly used for training embodied agents in real-time simulators such as Omniverse and Unreal. The system provides a constraint-based arrangement system that enables users to specify scene arrangement constraints, including symmetry, spatial relations, quantity, physics, and accessibility. The system also includes a constraint solver that searches for arrangements that maximally satisfy the constraints. Infinigen Indoors generates diverse indoor objects and full indoor scenes with physically and semantically plausible arrangements. It is fully procedural, using no external assets and only mathematical rules to generate everything from scratch. The system includes a library of procedural generators of indoor assets, a constraint-based arrangement system, and a tool to export generated scenes to real-time simulators. The system is designed to be highly customizable and allows users to specify high-level goals for abstract classes of objects. The system is also capable of generating realistic indoor scenes with diverse room combinations, connectedness, and overall contours. The system has been validated through experiments on tasks such as shadow removal and occlusion boundary estimation, demonstrating its effectiveness in generating synthetic data for training embodied agents. Infinigen Indoors is open-sourced under the BSD license to enable free and unlimited use by everyone and to enable community contributions of additional procedural generators.
Reach us at info@study.space
Understanding Infinigen Indoors%3A Photorealistic Indoor Scenes using Procedural Generation