The paper presents a method for interpreting indoor scenes from RGBD images, focusing on segmenting surfaces, objects, and inferring support relationships. The authors aim to handle typical, often messy, indoor scenes and introduce a novel integer programming formulation to infer physical support relations. They also provide a large dataset of 1449 RGBD images with detailed annotations, capturing 464 diverse indoor scenes. The approach involves inferring the 3D structure of the scene, segmenting it into objects and surfaces, and estimating support relations using both image and depth cues. The method uses structural classes to aid segmentation and support estimation, and it is robust to clutter, occlusion, and invisible supporting surfaces. The paper includes experiments demonstrating the effectiveness of the method in complex scenes and the benefits of 3D scene cues and inferred support for better object segmentation.The paper presents a method for interpreting indoor scenes from RGBD images, focusing on segmenting surfaces, objects, and inferring support relationships. The authors aim to handle typical, often messy, indoor scenes and introduce a novel integer programming formulation to infer physical support relations. They also provide a large dataset of 1449 RGBD images with detailed annotations, capturing 464 diverse indoor scenes. The approach involves inferring the 3D structure of the scene, segmenting it into objects and surfaces, and estimating support relations using both image and depth cues. The method uses structural classes to aid segmentation and support estimation, and it is robust to clutter, occlusion, and invisible supporting surfaces. The paper includes experiments demonstrating the effectiveness of the method in complex scenes and the benefits of 3D scene cues and inferred support for better object segmentation.