Physically-Based Rendering for Indoor Scene Understanding Using Convolutional Neural Networks

TLDR

Indoor scene understanding, vital for robot navigation and human assistance, depends on scarce per‑pixel ground truth data, leading to synthetic data use, yet a systematic study of its generation is missing. We introduce a large‑scale synthetic dataset of 500 K physically‑based rendered images from 45 K realistic 3‑D indoor scenes. We evaluate how rendering techniques and lighting conditions influence training for surface normal prediction, semantic segmentation, and object boundary detection. Our results show that more realistic rendering improves performance and that pretraining with the dataset surpasses current state‑of‑the‑art on all three tasks.

Abstract

Indoor scene understanding is central to applications such as robot navigation and human companion assistance. Over the last years, data-driven deep neural networks have outperformed many traditional approaches thanks to their representation learning capabilities. One of the bottlenecks in training for better representations is the amount of available per-pixel ground truth data that is required for core scene understanding tasks such as semantic segmentation, normal prediction, and object boundary detection. To address this problem, a number of works proposed using synthetic data. However, a systematic study of how such synthetic data is generated is missing. In this work, we introduce a large-scale synthetic dataset with 500K physically-based rendered images from 45K realistic 3D indoor scenes. We study the effects of rendering methods and scene lighting on training for three computer vision tasks: surface normal prediction, semantic segmentation, and object boundary detection. This study provides insights into the best practices for training with synthetic data (more realistic rendering is worth it) and shows that pretraining with our new synthetic dataset can improve results beyond the current state of the art on all three tasks.