… deep structured model for 3D scene flow … scene can be composed by estimating the 3D
rigid motion of each actor. We first exploit deep learning to extract visual cues for each instance. …

… on advances in deep learning on unordered point sets [52, 82, 1]. Early methods for 3D …
avoid flow-level or instance-level supervision altogether and only assume the availability of …

… of scene flow: given a pair of stereo or RGB-D video frames, estimate pixelwise 3D motion.
We introduce RAFT-3D, a new deep architecture for scene flow. … or instance segmentation …

… Additionally, we use a local replacement strategy, motivated by proposal instances for
which depth and motion errors are not correlated. We posit that these largely result from the 2D …

… – during scene flow estimation. Since a larger support is required to fit a plane and its rigid
motion … Our aim is to estimate depth and a 3D scene flow vector for each pixel of the reference …

… 1 (b) shows an example of the estimated pseudo scene flow labels based on point matching
… Dealing with non-rigid motions and occlusion issues is worthy of in-depth exploration in our …

… A brief example of our results is shown in Figure 1. Following the methods of unsupervised …
the-shelf depth map with rigidity as a prior. We show that the depth and flow can be estimated …

… system, where both scene flow and depth are estimated from … the depth given by the sensor
may be used directly for scene … For example, a 3D rotation of a rigid surface induces a great …

… deep scene flow estimation on point clouds by combining the FlowNet architecture [9, 18]
with advances in deep … annotated with scene flow vectors and instance segmentation masks. …

… to constrain the non-rigid scene flow as rigid within a specific range… Lastly, our deep scene
flow prior captures a continuous flow … Figure 4: Qualitative example of a scene flow estimation …