Holistic indoor scene understanding by context-supported instance segmentation

We propose a new method flow that utilizes pixel-level labeling information for instance-level object detection in indoor scenes from RGB-D data. Semantic and instance segmentation are two different paradigms scene understanding usually accomplished separately independently. interested integrating the tasks synergistic way order to take advantage of their complementary nature comprehensive understanding. Our work can capitalize on any deep learning networks used semantic by treating intermediate layer as category-wise local output, which is optimized jointly considering both spatial fitness relational context encoded three graphical models, namely, vertical placement model (VPM), horizontal (HPM) non-placement (NPM). VPM, HPM NPM represent common but distinct configurations: vertical, hanging relationships, respectively. Experimental results standard datasets show our significantly improve small with promising overall performance competitive state-of-the-art methods.