Editable and Applicable Indoor Scene Rearrangement via Dynamic Programming

Scene synthesis is a technology that creates manipulated scenes based on a virtual environment or reality, enhancing visual immersion and facilitating diverse experiences across various domains. This study proposes a new method for arranging objects in an editable structure within indoor scenes using dynamic programming. The proposed method undergoes evaluation and analysis, considering practical application methods and refining synthesis outcomes. The optimal object arrangement approach for indoor scene synthesis via dynamic programming entails a blend of Top-Down problem-solving, wherein objects are interpreted individually based on the scenes' overarching structure and relationships (prior or pairwise) derived from sample scenes, and Bottom-Up problem-solving, which interprets the scenes' structure from the sub-relationships among object units (accessibility). Leveraging a cost function for each method, fast and accurate scene synthesis is achieved through an object arrangement process comprising selection, exploration, convergence, and move. To assess the performance and efficiency of the proposed method, scene synthesis outcomes were generated for eight distinct configurations, with average time and accuracy during the arrangement process evaluated. Furthermore, we verified the flexible editing process of object arrangement by considering user awareness and interaction. Additionally, comparative experiments with existing optimal solution techniques and deep learning-based methods substantiate that the proposed method synthesizes various indoor scenes more efficiently and accurately. Lastly, a user study targeting expert groups was conducted to validate the practical applicability of the proposed method in interior design and home styling.