This research addresses the complex challenge of object placement in images through an innovative deep learning-based approach that leverages advanced scene understan‎ding an‎d semantic processing capabilities. The proposed framework, GraPLUS, presents a unique method for context-aware positioning by intelligently integrating deep learning techniques with structural scene representation. Pretrained language models are utilized to transform scene information into rich semantic representations, enabling nuanced understan‎ding of descriptive features an‎d spatial patterns of objects. This facilitates precise interpretation of complex relationships among scene elements an‎d placement patterns. By employing advanced deep architectures an‎d multimodal attenThis research addresses the complex challenge of object placement in images through an innovative deep learning-based approach that leverages advanced scene understan‎ding an‎d semantiction mechanisms, the framework overcomes limitations of previous methods. Experimental results demonstrate that GraPLUS achieves a placement accuracy of 92.1% an‎d an FID score of 28.83 on the OPA dataset, outperforming existing state-of-the-art methods by 8.3%. In human eva‎luation studies involving 38 participants an‎d 964 samples, the proposed method outperformed other approaches in 51.8% of the cases. Key innovations of this research include the design of a deep architecture for structural scene information processing, the use of pretrained models for cross-domain knowledge transfer, the development of specialized attention mechanisms for object placement, an‎d the creation of a unified framework for integrating semantic an‎d geometric information. Extensive experiments highlight the superiority of GraPLUS in accurate object placement with a particular strength in preserving geometric consistency an‎d semantic relationships across diverse scenes.

مهران صفاياني

عبدالرضا ميرزايي

نادر كريمي , مازيار پالهنگ