Pixel-level Encoding and Depth Layering for Instance-level Semantic Labeling
This provides improved instance segmentation for street scene analysis, though it appears incremental as it builds on existing FCN methods with added depth and encoding.
The paper tackles instance-aware semantic labeling by using a fully convolutional network to predict semantic labels, depth, and instance center direction, then applying low-level vision techniques to achieve state-of-the-art instance segmentation on KITTI and Cityscapes datasets, outperforming existing works by a large margin.
Recent approaches for instance-aware semantic labeling have augmented convolutional neural networks (CNNs) with complex multi-task architectures or computationally expensive graphical models. We present a method that leverages a fully convolutional network (FCN) to predict semantic labels, depth and an instance-based encoding using each pixel's direction towards its corresponding instance center. Subsequently, we apply low-level computer vision techniques to generate state-of-the-art instance segmentation on the street scene datasets KITTI and Cityscapes. Our approach outperforms existing works by a large margin and can additionally predict absolute distances of individual instances from a monocular image as well as a pixel-level semantic labeling.