Efficient Joint Segmentation, Occlusion Labeling, Stereo and Flow Estimation

摘要: In this paper we propose a slanted plane model for jointly recovering an image segmentation, dense depth estimate as well boundary labels (such occlusion boundaries) from static scene given two frames of stereo pair captured moving vehicle. Towards goal new optimization algorithm our SLIC-like objective which preserves connecteness segments and exploits shape regularization in the form length. We demonstrate performance approach challenging flow KITTI benchmarks show superior results to state-of-the-art. Importantly, these can be achieved order magnitude faster than competing approaches.