Contextual Scene Augmentation and Synthesis via GSACNet

Abstract

Indoor scene augmentation has become an emerging topic in the field of computer vision and graphics with applications in augmented and virtual reality. However, current state-of-the-art systems using deep neural networks require large datasets for training. In this paper we introduce GSACNet, a contextual scene augmentation system that can be trained with limited scene priors. GSACNet utilizes a novel parametric data augmentation method combined with a Graph Attention and Siamese network architecture followed by an Autoencoder network to facilitate training with small datasets. We show the effectiveness of our proposed system by conducting ablation and comparative studies with alternative systems on the Matterport3D dataset. Our results indicate that our scene augmentation outperforms prior art in scene synthesis with limited scene priors available.

arXiv commentarXiv admin note: text overlap with arXiv:2009.12395 by other authors

References in corpus (10)

• Semi-Supervised Classification with Graph Convolutional Networks
• Fast Graph Representation Learning with PyTorch Geometric
• ScanNet: Richly-annotated 3D Reconstructions of Indoor Scenes
• AI2-THOR: An Interactive 3D Environment for Visual AI
• Link Prediction Based on Graph Neural Networks
• Learning to Measure Change: Fully Convolutional Siamese Metric Networks for Scene Change Detection
• Interpolated Convolutional Networks for 3D Point Cloud Understanding
• OpenRooms: An End-to-End Open Framework for Photorealistic Indoor Scene Datasets
• SceneGen: Generative Contextual Scene Augmentation using Scene Graph Priors
• SceneSuggest: Context-driven 3D Scene Design