Pith. sign in

REVIEW 3 major objections 4 minor 44 references

Flash3D: Super-scaling Point Transformers through Joint Hardware-Geometry Locality

T0 review · 3 major / 4 minor · reviewed 2026-08-11 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Flash3D claims that aligning point-cloud bucketing with GPU tile boundaries lets Swin-style attention shifts run at zero extra cost, beating the prior state of the art by 2.25x in speed and 2.4x in memory.

desk verdict Clever bucket-aligned attention trick with a plausible zero-overhead argument; the speed claims hold up analytically, but the accuracy gains over PTv3 rest on undocumented baselines and no code. read the letter →

arxiv 2412.16481 v1 pith:PXQQCMDH submitted 2024-12-21 cs.CV

Liyan Chen , Gregory P. Meyer , Zaiwei Zhang , Eric M. Wolff , Paul Vernaza This is my paper · ORCID
classification cs.CV
keywords pointcloudtransformer3DsemanticsegmentationPerfectSpatialHashingFlashAttentionGPUtilingSwinefficientscalablebackbone
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Flash3D Transformer claims that the main obstacle to scaling point-cloud transformers is not attention itself but the global shuffling that reorders points between attention rounds, and that this obstacle can be removed by aligning geometric locality with GPU tiling. The paper's Perfect Spatial Hashing scatters points once into contiguous buckets where memory adjacency implies spatial adjacency; afterward, Swin-style window shifts and strides are implemented by redirecting which bucket tiles FlashAttention-2 loads, so no bytes are moved. On nuScenes semantic segmentation, the paper reports 81.2 mIoU at 20 ms and 0.5 G memory versus PTv3's 80.4 mIoU at 45 ms and 1.2 G with equal parameters, and 81.5 mIoU with the same memory budget at 2.8x the parameters. If the zero-cost shift claim holds, the result is a point-cloud backbone whose speed and memory scale with GPU tile schedules rather than with serialization cost, enabling wider attention scopes and larger models at the same budget.

What carries the argument

The central object is the bucket: a fixed-size, memory-contiguous group of points produced by Perfect Spatial Hashing (PSH), a bijection that maps spatially sparse 3D points into a compact array so that memory-address proximity equals spatial proximity. All bucket sizes are multiples of 16, so each bucket-aligned subarray is exactly a GPU tile for FlashAttention-2's 16x16 tile operations. Bucket-Swin attention defines an attention scope as a logical set of buckets and shifts or strides that set by changing which tile addresses the kernel fetches into L1, avoiding any permutation of the feature array. The mechanism carries the argument by replacing the global shuffle of previous point transformers with one up-front PSH scatter plus local kernel-level address redirection, and by confining pooling to the same tiles.

What would settle it

Profile the fused attention kernel with hardware counters while comparing three cases—contiguous buckets, shifted buckets, and strided buckets at the same attention scope size—and check whether the shifted or strided cases show measurably more tile loads, cache misses, or DRAM transactions than the contiguous case; if they do, the zero-overhead Bucket-Swin claim is false and the speedup must be recomputed.

Watch

Extended reading notes

Core claim

On its own terms, the paper establishes a claim about data movement: point-cloud attention should be organized so that every attention scope is a logical set of contiguous, bucket-aligned subarrays that fit GPU tiles, instead of physically re-serializing the point cloud for each new neighborhood definition. The discovery is that FlashAttention-2's tile addressing can be redirected to arbitrary bucket-aligned subarrays, making shifted and strided windows a zero-extra-cost operation after one initial Perfect Spatial Hashing scatter. This turns the expensive global shuffle of prior point transformers into a negligible one-time cost (the paper reports PSH latency at 0.19% of total latency versus serialization dominating PTv3), and it lets pooling run inside GPU tiles. The claimed consequence is state-of-the-art accuracy with a 2.25x speedup and 2.4x memory reduction over PTv3 on nuScenes, and the same pattern on Waymo.

Load-bearing premise

The speed gain depends on the assumption that redirecting which fixed-size data tiles the attention kernel loads, so that attention scopes span non-contiguous buckets, adds no extra memory traffic or latency beyond a plain contiguous-tile attention pass; if that redirection is not free, the reported 2.25x speedup weakens.

Editorial extensions

If this is right

  • If the central claim is correct, point-cloud transformers can drop global sorting from their forward pass entirely; window shifts and strides become address redirection inside the attention kernel.
  • Attention scopes can be widened (e.g., to 8192 points) and model width expanded under a fixed memory budget without the usual scaling penalty, which the paper shows as rising mIoU on nuScenes and Waymo when memory is held constant.
  • Equal-parameter comparisons on nuScenes (81.2 vs. 80.4 mIoU at 2.25x speed and 2.4x less memory) and equal-memory comparisons (81.5 mIoU with 2.8x parameters) are concrete corollaries of the design, not tuning artifacts, if the speed claim holds.
  • In-bucket pooling removes the need for global neighbor queries during downsampling, so backbone throughput should stay roughly linear in input size rather than super-linear, matching the paper's latency scaling profile.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Editorial inference: the same bucket-tile alignment should transfer to other sparse-data transformers on tiled GPUs (graphs, masked images, voxel grids), since the speedup mechanism is about replacing global reordering with tile-address redirection, not about 3D data specifically.
  • Editorial inference: the four hand-defined hash functions leave room for learned or data-dependent hash selection; if locality quality is what drives the accuracy gains, an optimized hash family could raise accuracy further without changing the kernel.
  • Editorial inference: PTv3's reported sub-5% TensorCore utilization suggests that many current point-cloud backbones are memory-movement-bound rather than compute-bound; if that diagnosis generalizes, Flash3D is a template for reworking other backbones, though the paper only demonstrates the comparison against PTv3.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. Flash3D is a point-cloud transformer backbone that unifies geometric locality with GPU memory tiling. It first scatters points into contiguous buckets via Perfect Spatial Hashing (Algorithms 1 and 2), then applies multi-head self-attention with FlashAttention-2, where Swin-style scope shifts are realized by loading different bucket-aligned tiles rather than physically shuffling features (Section 4.2 and Appendix 9.3). In-bucket pooling reduces features with fixed reduction factors. The paper claims state-of-the-art accuracy on nuScenes and Waymo semantic segmentation while being 2.25x faster and 2.4x more memory-efficient than PTv3, and it provides latency and hardware-utilization profiling on A100 and H100 GPUs. The central claims are therefore both architectural (zero-overhead locality alignment) and empirical (accuracy and speed superiority over PTv3).

Significance. If the claims hold, this is a valuable systems contribution: it replaces PTv3's global serialization and shuffle step with a cheap PSH construction and a zero-overhead logical shift, and the reported 2.25x speedup and 2.4x memory reduction at equal parameters would be a strong efficiency result for point transformers. The paper includes useful ablations (Table 2) and hardware profiling (Figures 5-13), and the zero-overhead argument in Section 9.3 is analytically plausible under the stated invariants. The main weakness is empirical: the accuracy advantage over PTv3 is small (0.4-1.4 mIoU) and is reported without error bars or a clear baseline protocol, and the code is withheld. The contribution is therefore significant but conditional on verification.

major comments (3)
  1. [Section 5.1, Tables 1-4] The central accuracy claim is not yet substantiated. Tables 1, 3, and 4 report Flash3D gains over PTv3 of 0.8-1.4 mIoU on nuScenes and 0.4-1.2 mIoU on Waymo, but the paper gives no error bars, no number of seeds, and no training recipe (epochs, optimizer, learning-rate schedule, augmentations, input sampling). Section 9 states that code will be released only after review, and there is no statement of whether the PTv3 numbers were re-run under the same protocol or taken from the original paper. Since Table 2 shows that individual design choices (hash combination, rebalancing, stride, swin) move mIoU by 0.1-1.0 points, the reported margins are within the range of protocol sensitivity. The 'outperforms PTv3' claim therefore needs either re-run baselines under identical conditions or seed-averaged statistics before it can be accepted.
  2. [Section 4.1, Algorithms 1-2; Section 9.3] The presented PSH algorithms do not enforce the fixed bucket capacity that the zero-overhead attention argument requires. In Algorithm 1, the check `if bkt ctr[hi] < S` followed by `ATOMIC INC` is a read-check-act race: two threads can both observe count S-1 and increment, producing S+1 points in a bucket. In Algorithm 2, the commit condition `prev_off <= S` accepts offset S, one past the valid range 0..S-1 for capacity S. Section 9.1's two-stage counters do not remove the race, because the bulk `AtomicAdd` commits the full local increment without a capacity check. If a bucket can exceed S, the feature array is no longer a concatenation of fixed-size, tile-aligned buckets, and Section 9.3's claim that `FP16[512:1024,d]` is exactly one bucket collapses. Please either replace the pseudocode with a capacity-enforcing parallel algorithm (e.g., atomic CAS or a two-phase scheme that checks the returned offset before commit) or state explicitly how the implementation guarantees the bucket-size invariant.
  3. [Section 9.3; Section 5.2] The 'zero-overhead' claim is supported only by an analytic argument, not by a direct measurement. The argument in Section 9.3 is plausible under the stated assumptions (contiguous concatenated buckets, bucket sizes multiples of 16, scopes formed by contiguous bucket ranges), and the latency breakdowns in Figures 5-7 are consistent with it. However, because the 2.25x speedup is a headline result, the paper should include a microbenchmark that isolates Bucket-Swin attention from plain FlashAttention-2 on the same layout, and should report the profiling setup (batch size, precision, PTv3 version, input generation) for Figures 8-13. Without this, the reader cannot distinguish the claimed zero-overhead switch from a small but real per-tile redirection cost.
minor comments (4)
  1. [Section 4.1] The sentence 'The index of the i-th point can be determined bybucket base[bucket id[i]]+ bucket offset[i]' is missing spaces and should reference the defined notation (e.g., bucket_base, bucket_id, bucket_offset); also 'geometric overage' in the XOR-mod description is likely 'geometric coverage'.
  2. [Figure 1 caption] The caption 'Effectiveness our Flash3D transformer...' is missing 'of'; please rephrase.
  3. [Table 2 caption] The header 'Hashs' should be 'Hashes', and the abbreviations XD, ZD, XM, ZM should be defined in the caption (XOR-div, Zorder-div, XOR-mod, Zorder-mod).
  4. [Section 5.1] The sentence 'Flash3D outperforms PTv3 on mIoU for 1.0% at 2.25x inference speed' should be 'by 1.0%' and should state the absolute mIoU gain (0.8 points) to avoid ambiguity; similarly, 'PTv3 uses 2.4x amount of memory than Flash3D' needs rewording.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation found; the central claims rest on external benchmarks and an independent architectural argument.

full rationale

The paper's derivation chain is self-contained against external evidence. The PSH-based bucketing is a construction algorithm with stated inputs and outputs, and the claim that contiguous memory proximity implies spatial proximity is a property of that construction, not a conclusion derived from the target performance numbers. The Bucket-Swin attention mechanism is justified by an independent architectural argument in Appendix 9.3: because feature arrays are concatenations of buckets whose sizes are multiples of 16, redirecting tile addresses changes which 16x16 K/V tiles are loaded without changing the number of tile loads. That argument may rest on an unverified hardware assumption about address-redirection costs, but it is not circular: the conclusion 'no overheads' is not assumed in the premise, and the paper does not define the conclusion into existence. The reported mIoU gains over PTv3 are empirical comparisons against external baseline numbers and datasets, not quantities derived from fitted parameters renamed as predictions. Validation-set ablations in Table 2 select hash-function combinations and attention-scope shifts, but reporting the resulting validation mIoU is standard evaluation rather than a fitted-input-called-prediction pattern. There are no load-bearing self-citations: the cited PSH, FlashAttention, ThunderKittens, and PTv3 works are all external to the authorship, and no uniqueness theorem or ansatz is imported from the authors' own prior work. The absence of error bars and training-protocol details for the PTv3 baseline is a legitimate correctness and reproducibility concern, but under the stated review rules it is not a circularity argument. No step in the paper reduces, by the paper's own equations or by self-citation, to its own inputs.

Assumptions & free parameters 6 free parameters · 5 assumptions · 0 invented entities

The central claims rest on several hand-set hyperparameters, including bucket capacity, hash selection, attention scope, and pooling factor, and on domain assumptions about GPU tiling and PSH locality. No new physical entities are introduced.

free parameters (6)
  • Bucket capacity S = Not stated exactly; example uses 512, pooling loads 1024 points per tile
    Defines bucket size and GPU tile alignment; set by hand and affects speed and locality.
  • Number of buckets K = 256 in the worked counter example
    Controls PSH table granularity; chosen for hardware and dataset scale.
  • Attention scope size = 4096 default (1024 and 8192 ablated)
    Core scaling hyperparameter; Table 5 shows task accuracy depends on it.
  • Hash function combination = XOR-div + XOR-mod + Zorder-div + Zorder-mod stacked
    Selected via validation ablation in Table 2; manually defined, not learned.
  • Pooling reduction factor rho = 2
    Fixed pooling ratio for in-bucket pooling described in Section 4.2.
  • Bucket stride and swin shift distances = Stride of 2 buckets, swin shift of 1 bucket
    Ablated in Table 2; set by hand to change receptive fields.
assumptions (5)
  • domain assumption Perfect spatial hashing yields a bijection that packs sparse 3D points into a contiguous array where address proximity implies spatial proximity.
    Lifted from graphics literature [1,17] and used as the basis of Section 4.1; not re-proven for LiDAR distributions.
  • domain assumption GPU architectures are tiled, with L1/shared memory and 16x16 matrix tiles, and the H100 L2 cache can coalesce the PSH global shuffle at near memory-bandwidth limits.
    Invoked in Section 3 and Appendix 9.1; hardware behavior from [12] and profiling.
  • domain assumption Redirecting FlashAttention-2 tile addresses so attention scopes include arbitrary bucket subarrays adds no extra memory traffic or latency beyond normal MHSA.
    Central to the zero-overhead Bucket-Swin claim in Section 4.2 and Appendix 9.3; not separately microbenchmarked.
  • domain assumption Optimistic racing and the recycle bucket in Algorithm 2 preserve enough geometric locality for downstream attention and pooling.
    Section 4.1; locality is best-effort and validated only empirically on nuScenes and Waymo.
  • domain assumption The four hand-designed hash functions (XOR-mod, XOR-div, Zorder-mod, Zorder-div) capture useful locality for point cloud attention.
    Section 4.1 and Table 2; selected by validation, not by a mathematical guarantee.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Flash3D: Super-scaling Point Transformers through Joint Hardware-Geometry Locality." pith.science (2026). https://pith.science/paper/PXQQCMDH

@misc{pith2026241216481,
  author       = {Pith},
  title        = {Pith review of: Flash3D: Super-scaling Point Transformers through Joint Hardware-Geometry Locality},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PXQQCMDH}},
  note         = {Machine review of arXiv:2412.16481}
}
read the original abstract

Recent efforts recognize the power of scale in 3D learning (e.g. PTv3) and attention mechanisms (e.g. FlashAttention). However, current point cloud backbones fail to holistically unify geometric locality, attention mechanisms, and GPU architectures in one view. In this paper, we introduce Flash3D Transformer, which aligns geometric locality and GPU tiling through a principled locality mechanism based on Perfect Spatial Hashing (PSH). The common alignment with GPU tiling naturally fuses our PSH locality mechanism with FlashAttention at negligible extra cost. This mechanism affords flexible design choices throughout the backbone that result in superior downstream task results. Flash3D outperforms state-of-the-art PTv3 results on benchmark datasets, delivering a 2.25x speed increase and 2.4x memory efficiency boost. This efficiency enables scaling to wider attention scopes and larger models without additional overhead. Such scaling allows Flash3D to achieve even higher task accuracies than PTv3 under the same compute budget.

Figures

Figures reproduced from arXiv: 2412.16481 by the authors.

Figure 1
Figure 1. Effectiveness our Flash3D transformer by unifying ge [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. High-level schematic overview of PTv3. Numbered rectangles represent locations in memory. Adjacent rectangles are adjacent [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. High-level schematic overview of Flash3D. Flash3D performs multiple rounds of attention with different neighborhood defini [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (9 more)
Figure 4
Figure 4. Figure 4: Illustration of bucket assignments using four hash functions after rebalancing. Colors of points indicate their bucket assignments. [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Overall Latencies vs. Input Sizes for Flash3D and PTv3. [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Latency TreeMap breakdowns for Flash3D and PTv3. [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 9
Figure 9. Figure 9: TensorCore Active Rates vs. Input Sizes for Flash3D [PITH_FULL_IMAGE:figures/full_fig_p008_9.png]
Figure 10
Figure 10. Figure 10: DRAM Read Bandwidth Usage vs. Input Sizes for [PITH_FULL_IMAGE:figures/full_fig_p008_10.png]
Figure 8
Figure 8. Figure 8: SM Utilization vs. Input Sizes for Flash3D and PTv3. [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]
Figure 12
Figure 12. Figure 12: TensorCore Active Rates vs. Input Sizes for Flash3D [PITH_FULL_IMAGE:figures/full_fig_p010_12.png]
Figure 11
Figure 11. Figure 11: SM Utilization vs. Input Sizes for Flash3D and PTv3. [PITH_FULL_IMAGE:figures/full_fig_p010_11.png]
Figure 13
Figure 13. Figure 13: DRAM Read Bandwidth Usage vs. Input Sizes for [PITH_FULL_IMAGE:figures/full_fig_p011_13.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

44 extracted references · 30 canonical work pages

  1. [1]

    Real-time parallel hashing on the gpu

    Dan A Alcantara, Andrei Sharf, Fatemeh Abbasinejad, Shubhabrata Sengupta, Michael Mitzenmacher, John D Owens, and Nina Amenta. Real-time parallel hashing on the gpu. In ACM SIGGRAPH asia 2009 papers, pages 1–9

    work page 2009

  2. [2]

    nuscenes: A multi- modal dataset for autonomous driving

    Holger Caesar, Varun Bankiti, Alex H Lang, Sourabh V ora, Venice Erin Liong, Qiang Xu, Anush Krishnan, Yu Pan, Gi- ancarlo Baldan, and Oscar Beijbom. nuscenes: A multi- modal dataset for autonomous driving. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 11621–11631, 2020. 6, 1

    work page 2020

  3. [3]

    2-s3net: Attentive feature fusion with adap- tive feature selection for sparse semantic segmentation net- work

    Ran Cheng, Ryan Razani, Ehsan Taghavi, Enxu Li, and Bingbing Liu. 2-s3net: Attentive feature fusion with adap- tive feature selection for sparse semantic segmentation net- work. In Proceedings of the IEEE/CVF conference on com- puter vision and pattern recognition , pages 12547–12556,

    work page

  4. [4]

    4d spatio-temporal convnets: Minkowski convolutional neural networks

    Christopher Choy, JunYoung Gwak, and Silvio Savarese. 4d spatio-temporal convnets: Minkowski convolutional neural networks. In Proceedings of the IEEE Conference on Com- puter Vision and Pattern Recognition , pages 3075–3084,

    work page

  5. [5]

    Perfect hashing

    Zbigniew J Czech, George Havas, and Bohdan S Majewski. Perfect hashing. Theoretical Computer Science , 182(1-2): 1–143, 1997. 2

    work page 1997

  6. [6]

    Flashattention-2: Faster attention with bet- ter parallelism and work partitioning

    Tri Dao. Flashattention-2: Faster attention with bet- ter parallelism and work partitioning. arXiv preprint arXiv:2307.08691, 2023. 1, 2, 3, 4

    arXiv 2023

  7. [7]

    Flashattention: Fast and memory-efficient exact at- tention with io-awareness

    Tri Dao, Dan Fu, Stefano Ermon, Atri Rudra, and Christo- pher R´e. Flashattention: Fast and memory-efficient exact at- tention with io-awareness. Advances in Neural Information Processing Systems, 35:16344–16359, 2022. 2, 3

    work page 2022

  8. [8]

    Embracing single stride 3d object detector with sparse trans- former

    Lue Fan, Ziqi Pang, Tianyuan Zhang, Yu-Xiong Wang, Hang Zhao, Feng Wang, Naiyan Wang, and Zhaoxiang Zhang. Embracing single stride 3d object detector with sparse trans- former. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 8458–8468,

Show all 44 references
  1. [9]

    Panoptic nuscenes: A large-scale benchmark for lidar panoptic segmentation and tracking

    Whye Kit Fong, Rohit Mohan, Juana Valeria Hurtado, Lub- ing Zhou, Holger Caesar, Oscar Beijbom, and Abhinav Val- ada. Panoptic nuscenes: A large-scale benchmark for lidar panoptic segmentation and tracking. IEEE Robotics and Au- tomation Letters, 7(2):3795–3802, 2022. 6, 1

    2022

  2. [10]

    Paterson, and Kyle Johnsen

    Anton Franzluebbers, Changying Li, Andrew H. Paterson, and Kyle Johnsen. Virtual reality point cloud annotation. 2022 IEEE Conference on Virtual Reality and 3D User In- terfaces Abstracts and Workshops (VRW) , pages 886–887,

    2022

  3. [11]

    Storing a sparse table with 0 (1) worst case access time.Jour- nal of the ACM (JACM), 31(3):538–544, 1984

    Michael L Fredman, J ´anos Koml ´os, and Endre Szemer ´edi. Storing a sparse table with 0 (1) worst case access time.Jour- nal of the ACM (JACM), 31(3):538–544, 1984. 2

    1984

  4. [12]

    Hennessy and David A

    John L. Hennessy and David A. Patterson. Computer Ar- chitecture, Sixth Edition: A Quantitative Approach. Morgan Kaufmann Publishers Inc., San Francisco, CA, USA, 6th edi- tion, 2017. 2 5

    2017

  5. [13]

    Processing, assessing, and enhancing the waymo autonomous vehicle open dataset for driving behavior re- search

    Xiangwang Hu, Zuduo Zheng, Danjue Chen, Xi Zhang, and Jian Sun. Processing, assessing, and enhancing the waymo autonomous vehicle open dataset for driving behavior re- search. Transportation Research Part C: Emerging Tech- nologies, 134:103490, 2022. 1

    2022

  6. [14]

    Planning-oriented autonomous driving

    Yihan Hu, Jiazhi Yang, Li Chen, Keyu Li, Chonghao Sima, Xizhou Zhu, Siqi Chai, Senyao Du, Tianwei Lin, Wenhai Wang, et al. Planning-oriented autonomous driving. In Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 17853–17862, 2023. 1

    2023

  7. [15]

    Stratified trans- former for 3d point cloud segmentation

    Xin Lai, Jianhui Liu, Li Jiang, Liwei Wang, Hengshuang Zhao, Shu Liu, Xiaojuan Qi, and Jiaya Jia. Stratified trans- former for 3d point cloud segmentation. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 8500–8509, 2022. 1

    2022

  8. [16]

    Spherical transformer for lidar-based 3d recognition

    Xin Lai, Yukang Chen, Fanbin Lu, Jianhui Liu, and Jiaya Jia. Spherical transformer for lidar-based 3d recognition. In Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 17545–17555, 2023. 1

    2023

  9. [17]

    Perfect spatial hash- ing

    Sylvain Lefebvre and Hugues Hoppe. Perfect spatial hash- ing. ACM Transactions on Graphics (TOG), 25(3):579–588,

  10. [18]

    Swin transformer: Hierarchical vision transformer using shifted windows

    Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, and Baining Guo. Swin transformer: Hierarchical vision transformer using shifted windows. In Proceedings of the IEEE/CVF international conference on computer vision, pages 10012–10022, 2021. 1, 2

    2021

  11. [19]

    Flatformer: Flattened window attention for efficient point cloud transformer

    Zhijian Liu, Xinyu Yang, Haotian Tang, Shang Yang, and Song Han. Flatformer: Flattened window attention for efficient point cloud transformer. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 1200–1211, 2023. 1, 2, 5

    2023

  12. [20]

    Spatial hash- joins

    Ming-Ling Lo and Chinya V Ravishankar. Spatial hash- joins. In Proceedings of the 1996 ACM SIGMOD interna- tional conference on Management of data , pages 247–258,

    1996

  13. [21]

    Point cloud based reinforcement learning for sim-to-real and partial ob- servability in visual navigation

    Kenzo Lobos-Tsunekawa and Tatsuya Harada. Point cloud based reinforcement learning for sim-to-real and partial ob- servability in visual navigation. 2020 IEEE/RSJ Interna- tional Conference on Intelligent Robots and Systems (IROS), pages 5871–5878, 2020. 1

    2020

  14. [22]

    Online normalizer calculation for softmax

    Maxim Milakov and Natalia Gimelshein. Online normalizer calculation for softmax. arXiv preprint arXiv:1805.02867 ,

    arXiv

  15. [23]

    A computer oriented geodetic data base and a new technique in file sequencing

    Guy M Morton. A computer oriented geodetic data base and a new technique in file sequencing. 1966. 5

    1966

  16. [24]

    Weiss, Benjamin Sapp, Zhifeng Chen, and Jonathon Shlens

    Jiquan Ngiam, Benjamin Caine, Vijay Vasudevan, Zheng- dong Zhang, Hao-Tien Lewis Chiang, Jeffrey Ling, Re- becca Roelofs, Alex Bewley, Chenxi Liu, Ashish Venugopal, David J. Weiss, Benjamin Sapp, Zhifeng Chen, and Jonathon Shlens. Scene transformer: A unified multi-task model ...

    arXiv

  17. [25]

    Nvidia nsight systems 2024.6.1

    CUDA Nvidia. Nvidia nsight systems 2024.6.1. NVIDIA: Santa Clara, CA, 2024. 7

    2024

  18. [26]

    Dexpoint: Generalizable point cloud rein- forcement learning for sim-to-real dexterous manipulation

    Yuzhe Qin, Binghao Huang, Zhao-Heng Yin, Hao Su, and Xiaolong Wang. Dexpoint: Generalizable point cloud rein- forcement learning for sim-to-real dexterous manipulation. In Conference on Robot Learning, 2022. 1

    2022

  19. [27]

    Semantic segmen- tation for real point cloud scenes via bilateral augmentation and adaptive fusion

    Shi Qiu, Saeed Anwar, and Nick Barnes. Semantic segmen- tation for real point cloud scenes via bilateral augmentation and adaptive fusion. 2021 IEEE/CVF Conference on Com- puter Vision and Pattern Recognition (CVPR), pages 1757– 1767, 2021. 1

    2021

  20. [28]

    Build- ingnet: Learning to label 3d buildings

    Pratheba Selvaraju, Mohamed Nabail, Marios Loizou, Maria Maslioukova, Melinos Averkiou, Andreas Andreou, Sid- dhartha Chaudhuri, and Evangelos Kalogerakis. Build- ingnet: Learning to label 3d buildings. In 2021 IEEE/CVF International Conference on Computer Vision, ICCV 2021, M...

    2021

  21. [29]

    Flashattention-3: Fast and accurate attention with asynchrony and low-precision

    Jay Shah, Ganesh Bikshandi, Ying Zhang, Vijay Thakkar, Pradeep Ramani, and Tri Dao. Flashattention-3: Fast and accurate attention with asynchrony and low-precision. arXiv preprint arXiv:2407.08608, 2024. 2, 3, 4

    2024 arXiv

  22. [30]

    Efficient attention: Attention with lin- ear complexities

    Zhuoran Shen, Mingyuan Zhang, Haiyu Zhao, Shuai Yi, and Hongsheng Li. Efficient attention: Attention with lin- ear complexities. In Proceedings of the IEEE/CVF winter conference on applications of computer vision, pages 3531– 3539, 2021. 2

    2021

  23. [31]

    Thunderkittens: Simple, fast, and adorable ai kernels

    Benjamin F Spector, Simran Arora, Aaryan Singhal, Daniel Y Fu, and Christopher R ´e. Thunderkittens: Simple, fast, and adorable ai kernels. arXiv preprint arXiv:2410.20399, 2024. 2, 6, 3

    2024 arXiv

  24. [32]

    Scalability in perception for autonomous driving: Waymo open dataset

    Pei Sun, Henrik Kretzschmar, Xerxes Dotiwalla, Aurelien Chouard, Vijaysai Patnaik, Paul Tsui, James Guo, Yin Zhou, Yuning Chai, Benjamin Caine, et al. Scalability in perception for autonomous driving: Waymo open dataset. In Proceed- ings of the IEEE/CVF conference on computer ...

    2020

  25. [33]

    Swformer: Sparse window transformer for 3d object detection in point clouds

    Pei Sun, Mingxing Tan, Weiyue Wang, Chenxi Liu, Fei Xia, Zhaoqi Leng, and Dragomir Anguelov. Swformer: Sparse window transformer for 3d object detection in point clouds. In European Conference on Computer Vision , pages 426–

  26. [34]

    Searching efficient 3d architec- tures with sparse point-voxel convolution

    Haotian Tang, Zhijian Liu, Shengyu Zhao, Yujun Lin, Ji Lin, Hanrui Wang, and Song Han. Searching efficient 3d architec- tures with sparse point-voxel convolution. In European con- ference on computer vision, pages 685–702. Springer, 2020. 1

    2020

  27. [35]

    Sim-to-real deep reinforcement learning for safe end-to-end planning of aerial robots

    Halil Ibrahim Ugurlu, Huy Xuan Pham, and Erdal Kayacan. Sim-to-real deep reinforcement learning for safe end-to-end planning of aerial robots. Robotics, 11:109, 2022. 1

    2022

  28. [36]

    Octformer: Octree-based transformers for 3d point clouds

    Peng-Shuai Wang. Octformer: Octree-based transformers for 3d point clouds. ACM Transactions on Graphics (TOG), 42(4):1–11, 2023. 1, 2

    2023

  29. [37]

    Point transformer v2: Grouped vector atten- tion and partition-based pooling

    Xiaoyang Wu, Yixing Lao, Li Jiang, Xihui Liu, and Heng- shuang Zhao. Point transformer v2: Grouped vector atten- tion and partition-based pooling. Advances in Neural Infor- mation Processing Systems, 35:33330–33342, 2022. 2, 5, 6, 1, 4

    2022

  30. [38]

    Point transformer v3: Simpler faster stronger

    Xiaoyang Wu, Li Jiang, Peng-Shuai Wang, Zhijian Liu, Xi- hui Liu, Yu Qiao, Wanli Ouyang, Tong He, and Hengshuang Zhao. Point transformer v3: Simpler faster stronger. In Pro- ceedings of the IEEE/CVF Conference on Computer Vision 6 and Pattern Recognition, pages 4840–4851, 2024...

    2024

  31. [39]

    Data- efficient learning for sim-to-real robotic grasping using deep point cloud prediction networks

    Xinchen Yan, Mohi Khansari, Jasmine Hsu, Yuanzheng Gong, Yunfei Bai, S ¨oren Pirk, and Honglak Lee. Data- efficient learning for sim-to-real robotic grasping using deep point cloud prediction networks. ArXiv, abs/1906.08989,

    1906 arXiv

  32. [40]

    Swin3d: A pretrained transformer backbone for 3d indoor scene understanding

    Yu-Qi Yang, Yu-Xiao Guo, Jian-Yu Xiong, Yang Liu, Hao Pan, Peng-Shuai Wang, Xin Tong, and Baining Guo. Swin3d: A pretrained transformer backbone for 3d indoor scene understanding. arXiv preprint arXiv:2304.06906 ,

    arXiv

  33. [41]

    Fusionad: Multi-modality fusion for pre- diction and planning tasks of autonomous driving

    Tengju Ye, Wei Jing, Chunyong Hu, Shikun Huang, Ling- ping Gao, Fangzhen Li, Jingke Wang, Ke Guo, Wencong Xiao, Wei Mao, Hang Zheng, Kun Li, Junbo Chen, and Kaicheng Yu. Fusionad: Multi-modality fusion for pre- diction and planning tasks of autonomous driving. ArXiv, abs/2308....

    2023 arXiv

  34. [42]

    Toward zero-shot sim-to-real transfer learning for pneumatic soft robot 3d proprioceptive sensing

    Uksang Yoo, Hanwen Zhao, Alvaro Altamirano, Wenzhen Yuan, and Chen Feng. Toward zero-shot sim-to-real transfer learning for pneumatic soft robot 3d proprioceptive sensing. 2023 IEEE International Conference on Robotics and Au- tomation (ICRA), pages 544–551, 2023. 1

    2023

  35. [43]

    Point transformer

    Hengshuang Zhao, Li Jiang, Jiaya Jia, Philip HS Torr, and Vladlen Koltun. Point transformer. In Proceedings of the IEEE/CVF international conference on computer vision, pages 16259–16268, 2021. 2

    2021

  36. [44]

    Cylinder3d: An effective 3d framework for driving-scene lidar semantic segmentation

    Hui Zhou, Xinge Zhu, Xiao Song, Yuexin Ma, Zhe Wang, Hongsheng Li, and Dahua Lin. Cylinder3d: An effective 3d framework for driving-scene lidar semantic segmentation. arXiv preprint arXiv:2008.01550, 2020. 1 7

    2008 arXiv

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.