A qualitative and localized evaluation for 3D indoor scene synthesis

We presented a novel evaluation method for assessing synthesized 3D indoor scenes. By exploiting the concept of local flaw identification in a navigable 3D viewer, this work was able to address the subjectivity, ambiguity, and low fidelity issues existing in conventional experiments. As a result, our method managed to produce a more consistent and reliable assessment outcome for a synthetic scene and to disclose meaningful information that will help scholars to investigate their research further. Two user studies were carried out to demonstrate and validate the effectiveness of our method.