Habitat Synthetic Scenes Dataset (HSSD-200): An Analysis of 3D Scene Scale and Realism Tradeoffs for ObjectGoal Navigation

Figure 1. Left: we contribute the Habitat Synthetic Scenes Dataset (HSSD-200), a new dataset of high-quality, human-authored synthetic 3D scenes. Right: zero-shot ObjectNav performance on HM3DSem [43] for agents pretained on synthetic 3D scene datasets of different scale and quality. Through a systematic analysis of scene dataset scale and realism our experiments show that the benefit of dataset scale saturates quickly, and scene realism and quality become the bottleneck for improved ObjectNav agent generalization to realistic scenes. Concretely, we find that agents trained on 122 scenes from HSSD outperform agents trained on two orders of magnitude more scenes from the ProcTHOR [14] dataset (19.2 vs 12.5 success rate).