More Like this

1. 3D SceneFlowNet: Self-supervised 3D Scene Flow Estimation based on Graph CNN

   https://doi.org/10.1109/ICIP42928.2021.9506286  

   Lu, Yawen; Zhu, Yuhao; Lu, Guoyu (January 2021, IEEE International Conference on Image Processing (ICIP))

   Despite deep learning approaches have achieved promising successes in 2D optical flow estimation, it is a challenge to accurately estimate scene flow in 3D space as point clouds are inherently lacking topological information. In this paper, we aim at handling the problem of self-supervised 3D scene flow estimation based on dynamic graph convolutional neural networks (GCNNs), namely 3D SceneFlowNet. To better learn geometric relationships among points, we introduce EdgeConv to learn multiple-level features in a pyramid from point clouds and a self-attention mechanism to apply the multi-level features to predict the final scene flow. Our trained model can efficiently process a pair of adjacent point clouds as input and predict a 3D scene flow accurately without any supervision. The proposed approach achieves superior performance on both synthetic ModelNet40 dataset and real LiDAR scans from KITTI Scene Flow 2015 datasets. 

   more » « less
2. 3D Simulation of an Extreme SAID Flow Channel

   https://doi.org/10.1029/2024JA032660  

   Díaz_Peña, Joaquín; Zettergren, Matthew; Semeter, Joshua; Nishimura, Yukitoshi; Hirsch, Michael; Walsh, Brian M (June 2024, Journal of Geophysical Research: Space Physics)

   Abstract Space‐based observations of the signatures associated with STEVE show how this phenomenon might be closely related to an extreme version of a SAID channel. Measurements show high velocities (>4 km/s), high temperatures (>4,000 K), and very large current density drivers (up to 1 μA/m²). This phenomena happens in a small range of latitudes, less than a degree, but with a large longitudinal span. In this study, we utilize the GEMINI model to simulate an extreme SAID/STEVE. We assume a FAC density coming from the magnetosphere as the main driver, allowing all other parameters to adjust accordingly. We have two main objectives with this work: show how an extreme SAID can have velocity values comparable or larger than the ones measured under STEVE, and to display the limitations and missing physics that arise due to the extreme values of temperature and velocity. Changes had to be made to GEMINI due to the extreme conditions, particularly some neutral‐collision frequencies. The importance of the temperature threshold at which some collision frequencies go outside their respective bounds, as well as significance of the energies that would cause inelastic collisions and impact ionization are displayed and discussed. We illustrate complex structures and behaviors, emphasizing the importance of 3D simulations in capturing these phenomena. Longitudinal structure is emphasized, as the channel develops differently depending on MLT. However, these simulations should be viewed as approximations due to the limited observations available to constrain the model inputs and the assumptions made to achieve sensible results. 

   more » « less
3. FlowNet3D: Learning Scene Flow in 3D Point Clouds

   Liu, Xingyu; Qi, Charles R.; Guibas, Leonidas J. (January 2019, CVPR)
4. FlowNet3D: Learning Scene Flow in 3D Point Clouds

   Liu, Xingyu; Qi, Charles R.; Guibas, Leonidas J. (January 2019, Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition)
5. RAFT-3D: Scene Flow using Rigid-Motion Embeddings

   https://doi.org/10.1109/CVPR46437.2021.00827  

   Teed, Zachary; Deng, Jia (January 2021, IEEE Conference on Computer Vision and Pattern Recognition (CVPR))