ASH: A Modern Framework for Parallel Spatial Hashing in 3D Perception

被引:5
作者
Dong, Wei [1 ]
Lao, Yixing [2 ]
Kaess, Michael [1 ,3 ]
Koltun, Vladlen [2 ,4 ]
机构
[1] Carnegie Mellon Univ, Robot Inst, Pittsburgh, PA 15213 USA
[2] Intel Corp, Intel Labs, Santa Clara, CA 95054 USA
[3] Georgia Tech, Coll Comp, Atlanta, GA 30332 USA
[4] Apple Inc, One Apple Pk Way, Cupertino, CA 95014 USA
关键词
Three-dimensional displays; Graphics processing units; Ash; Tensors; Task analysis; Image reconstruction; Arrays; Parallel hashing; GPU; volumetric reconstruction; SLAM; shape-from-shading; autodiff; RECONSTRUCTION;
D O I
10.1109/TPAMI.2022.3214347
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
We present ASH, a modern and high-performance framework for parallel spatial hashing on GPU. Compared to existing GPU hash map implementations, ASH achieves higher performance, supports richer functionality, and requires fewer lines of code (LoC) when used for implementing spatially varying operations from volumetric geometry reconstruction to differentiable appearance reconstruction. Unlike existing GPU hash maps, the ASH framework provides a versatile tensor interface, hiding low-level details from the users. In addition, by decoupling the internal hashing data structures and key-value data in buffers, we offer direct access to spatially varying data via indices, enabling seamless integration to modern libraries such as PyTorch. To achieve this, we 1) detach stored key-value data from the low-level hash map implementation; 2) bridge the pointer-first low level data structures to index-first high-level tensor interfaces via an index heap; 3) adapt both generic and non-generic integer-only hash map implementations as backends to operate on multi-dimensional keys. We first profile our hash map against state-of-the-art hash maps on synthetic data to show the performance gain from this architecture. We then show that ASH can consistently achieve higher performance on various large-scale 3D perception tasks with fewer LoC by showcasing several applications, including 1) point cloud voxelization, 2) retargetable volumetric scene reconstruction, 3) non-rigid point cloud registration and volumetric deformation, and 4) spatially varying geometry and appearance refinement. ASH and its example applications are open sourced in Open3D (http://www.open3d.org).
引用
收藏
页码:5417 / 5435
页数:19
相关论文
共 64 条
  • [41] KinectFusion: Real-Time Dense Surface Mapping and Tracking
    Newcombe, Richard A.
    Izadi, Shahram
    Hilliges, Otmar
    Molyneaux, David
    Kim, David
    Davison, Andrew J.
    Kohli, Pushmeet
    Shotton, Jamie
    Hodges, Steve
    Fitzgibbon, Andrew
    [J]. 2011 10TH IEEE INTERNATIONAL SYMPOSIUM ON MIXED AND AUGMENTED REALITY (ISMAR), 2011, : 127 - 136
  • [42] Niessner M, 2013, ACM T GRAPHIC, V32, P1, DOI DOI 10.1145/2508363.2508374
  • [43] Mitsuba 2: A Retargetable Forward and Inverse Renderer
    Nimier-David, Merlin
    Vicini, Delio
    Zeltner, Tizian
    Jakob, Wenzel
    [J]. ACM TRANSACTIONS ON GRAPHICS, 2019, 38 (06):
  • [44] Oleynikova H, 2017, IEEE INT C INT ROBOT, P1366
  • [45] Cuckoo hashing
    Pagh, R
    Rodler, FF
    [J]. JOURNAL OF ALGORITHMS-COGNITION INFORMATICS AND LOGIC, 2004, 51 (02): : 122 - 144
  • [46] Colored Point Cloud Registration Revisited
    Park, Jaesik
    Zhou, Qian-Yi
    Koltun, Vladlen
    [J]. 2017 IEEE INTERNATIONAL CONFERENCE ON COMPUTER VISION (ICCV), 2017, : 143 - 152
  • [47] Paszke A, 2019, ADV NEUR IN, V32
  • [48] Prisacariu VA, 2017, Arxiv, DOI arXiv:1708.00783
  • [49] R. D. Team, 2018, RAP COLL LIBR END TO
  • [50] Reiser S., 2021, ARXIV