Pixel Difference Convolutional Network for RGB-D Semantic Segmentation

RGB-D semantic segmentation can be advanced with convolutional neural networks due to the availability of Depth data. Although objects cannot easily discriminated by just 2D appearance, local pixel difference and geometric patterns in Depth, they well separated some cases. Considering fixed grid kernel structure, CNNs are limited lack ability capture detailed, fine-grained information thus achieve accurate pixel-level segmentation. To solve this problem CNN we propose a Pixel Difference Convolutional Network (PDCNet) detailed intrinsic aggregating both intensity gradient range for data global RGB data, respectively. Precisely, PDCNet consists branch an branch. For branch, Convolution (PDC) consider via information. contribute lightweight Cascade Large Kernel (CLK) extend PDC, namely CPDC, enjoy contexts further boost performance. Consequently, differences from modal seamlessly incorporated into during propagation process. Experiments on three challenging benchmark datasets, i.e ., NYUDv2 (78.4 Acc., 53.5 mIoU), SUN (83.3 49.6 mIoU) SID Dataset (83.1 61.4 reveal that our achieves state-of-the-art performance task.