Pith. sign in

REVIEW 5 major objections 5 minor 83 references

Self-Supervised Sparse Sensor Fusion for Long Range Perception

T0 review · 5 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read This paper claims that a fully sparse, self-supervised camera-LiDAR fusion architecture can extend 3D object detection and LiDAR forecasting to 250 m, outperforming BEV-based methods that scale quadratically with range.

desk verdict A solid fusion architecture with genuine NuScenes forecasting gains, but the paper's central 250m detection claim is not supported by the reported aggregate metrics alone. read the letter →

arxiv 2508.13995 v1 pith:KNQUKGUX submitted 2025-08-19 cs.CV

Edoardo Palladin , Samuel Brucker , Filippo Ghilotti , Praveen Narayanan , Mario Bijelic , Felix Heide This is my paper
classification cs.CV
keywords long-rangeperceptionsparsevoxelrepresentationLiDAR-camerafusionself-supervisedpretrainingDARforecasting3Dobjectdetectionhighwayautonomytruckingdataset
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

Long-range highway driving requires perception to at least 250 m, roughly five times the urban range of current BEV-based detectors. The paper argues that the standard dense BEV representation is the bottleneck: its memory and compute grow quadratically with range, and long-range objects are too sparse to label at scale. It proposes LRS4Fusion, a fully sparse voxel representation that fuses camera and LiDAR features at multiple scales and adds temporal context through a sparse windowed attention over past frames. A self-supervised pre-training stage learns occupancy and velocity by forecasting future LiDAR from raw, unlabeled sensor data. On a newly captured trucking dataset, the method reports 52.61 mAP at up to 250 m, +11.06 over the second-best method, and reduces Chamfer Distance in LiDAR forecasting by up to 30.5%.

What carries the argument

The central object is a fully sparse multi-scale voxel representation that stays sparse end-to-end. Camera features are projected into 3D through a lightweight recurrent depth-completion module with a minimal gated unit, concatenated with voxelized LiDAR features, and refined by sparse completion and context aggregation blocks. Temporal context comes from a sparse windowed attention layer in which each occupied current voxel queries occupied voxels in a warped past frame, avoiding any densification. Self-supervision is carried by sparse occupancy and velocity decoders: a query point in spacetime is interpolated from voxel features, moved by a learned pose network, and supervised by raw LiDAR

What would settle it

Recompute detection mAP on a held-out long-range subset whose 250-400 m labels are independently verified, after retraining BEVFusion, SAMFusion, and BEVFormer with the same long-range voxel sizes, history length, and compute budget. If the mAP gap to SAMFusion shrinks to near zero, the claimed long-range advantage is a protocol effect; if the gap persists, the sparse-representation claim is supported.

Watch

Extended reading notes

Core claim

LRS4Fusion establishes that long-range 3D perception does not require dense BEV grids. It keeps every feature stage sparse: camera features are lifted into 3D via learned dense depth maps refined by sparse LiDAR returns, concatenated with voxelized LiDAR features, processed by sparse completion and context aggregation blocks at four scales, and fused across time by letting each occupied current voxel attend to a window of occupied voxels from the previous frame warped by odometry and measured radial velocity. Before any labeled detection training, the model is pre-trained to reconstruct future occupancy and velocity from raw LiDAR rays, giving spatiotemporal understanding without manual labe

Load-bearing premise

The headline results rest on the assumption that the new private trucking dataset's 250-400 m annotations are complete and accurate, and that the compared baselines were retrained under the same 250 m protocol rather than handicapped by their default urban range settings.

Editorial extensions

If this is right

  • Object detection and LiDAR forecasting at 250 m become feasible without the quadratic memory cost of BEV grids, and the same architecture could extend further as sensor range grows.
  • Self-supervised pre-training on raw camera-LiDAR data replaces much of the labeled-data need: occupancy, velocity, depth, and forecasting are learned from unlabeled recordings, with manual annotations needed only for the final detection head.
  • LiDAR forecasting at 3-second horizons improves by up to roughly 29-30% over prior state of the art, giving planners a dynamic forecast of the scene rather than a static snapshot.
  • History length matters differently on highways: 1 second of history beats 3 seconds for detection because fast-moving objects leave the region of interest, revising an urban-centered design choice.
  • The multi-scale sparse voxel design plus a lightweight recurrent depth module reduces depth-completion runtime to 63 ms and memory to 1.3 GB while improving accuracy, making camera-to-3D lifting practical at long range.

Reading between the lines

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

  • The paper's own ablations attribute only part of the headline gain to self-supervision: the same architecture without it scores 49.58 mAP versus 52.61 with it. The remaining gap over SAMFusion therefore comes from the sparse representation, depth module, and backbone choices, an attribution the paper does not spell out.
  • If the long-range dataset is released, a natural test is whether the same self-supervised pre-training transfers to urban detection benchmarks; the paper reports NuScenes forecasting but not NuScenes object detection, leaving cross-dataset detection transfer open.
  • Because temporal fusion is bounded by occupied voxels and never densifies, the method should extend beyond 250 m toward the sensor's full 400 m range at roughly linear cost, but the paper evaluates detection only up to 250 m.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 5 minor

Summary. The paper proposes LRS4Fusion, a fully sparse voxel-based camera-LiDAR fusion architecture for long-range (up to 250 m) 3D perception. The method fuses camera features lifted by a learned depth network with LiDAR voxel features, applies a sparse windowed temporal attention over past frames, and is pre-trained with self-supervised occupancy and velocity reconstruction losses before supervised object detection. Claims are: (i) state-of-the-art 3D object detection on a new private long-range trucking dataset, with +11.06 mAP / 26.6% over the second-best method; (ii) state-of-the-art LiDAR forecasting on the same dataset, with up to 30.5% Chamfer Distance improvement and 29% improvement on NuScenes; and (iii) an efficient depth completion module with 1.3 GB memory and 64 ms runtime. The evaluation combines a private dataset with NuScenes as an external benchmark.

Significance. If the long-range claims hold, the work is significant: it demonstrates that a fully sparse multi-modal voxel representation with temporal fusion and self-supervised pretraining can scale perception beyond the usual 50–100 m urban operating range, which is practically important for highway and trucking autonomy. The use of a private dataset with boxes up to 400 m is a valuable contribution, and the proposed efficiency gains in depth completion (Table 1) are concrete and reproducible in principle. The NuScenes forecasting results provide an external anchor that supports the transferability of the forecasting pre-training. However, the headline long-range claims are supported only by aggregate metrics over a close-range-dominated test distribution, and key baseline adaptation details are deferred to the supplement. The paper is therefore a promising but not yet fully verified contribution.

major comments (5)
  1. [Sec. 5.2, Table 2] The central claim that the method 'extends 3D object detection to 250 m' rests entirely on aggregate mAP/NDS over the full test set. Fig. 2 shows the instance distribution is heavily concentrated at close range (compared to NuScenes). Without range-resolved evaluation (e.g., AP in 0–50 m, 50–100 m, 100–150 m, 150–200 m, 200–250 m bins, or at least a separate AP for objects beyond 100 m), the +11.06 mAP improvement could be driven entirely by close-range objects. Range-binned AP is standard in long-range perception (e.g., Waymo). Please report range-stratified AP; if the improvement is not present beyond 100 m, the 250 m claim should be revised.
  2. [Sec. 5.2, Table 2] The table shows LRS4Fusion (w/o Pre-train) with NDS 59.12 and LRS4Fusion (with pre-train) with NDS 58.06, while mAP improves from 49.58 to 52.61. The manuscript states in Sec. 5.2 that 'the proposed occupancy-velocity self-supervision improves performance' and Table 7(c) reports a +6.11% mAP improvement. The NDS drop is not discussed. Since NDS is a headline metric, this inconsistency must be explained (e.g., metric instability, different training runs, or a trade-off between mAP and NDS). If NDS is not a primary metric, state this explicitly.
  3. [Sec. 5.2, Table 2 and Sec. 4] The comparison against baselines (PointPillars, BEVFormer, BEVFusion, SAMFusion) on the private long-range dataset is unverifiable from the main text. These methods are designed for 50–100 m ranges; the paper does not state whether they were retrained on the long-range dataset with extended voxel ranges, whether the same ROI and anchor assignments were used, or what sensor inputs (e.g., 5 Hz vs. 10 Hz, 400 m vs. 100 m LiDAR) were provided. Baseline adaptation is deferred to the supplementary material. If the baselines were not adapted to the 250 m regime under a comparable protocol, the reported +26.6% mAP advantage is not meaningful. Please include the baseline adaptation protocol in the paper, or make the supplement available for review.
  4. [Sec. 5.3, Table 3] LiDAR forecasting Chamfer Distance is reported as a single scalar over the full ROI (x ∈ [−100, 250] m, y ∈ [−100, 100] m). Point density and geometric complexity are much higher at close range, so a method that predicts near-field geometry well can achieve low aggregate CD without solving long-range forecasting. To support the 'up to 250 m' claim, report CD in distance slices (e.g., 0–50 m, 50–100 m, 100–150 m, 150–200 m, 200–250 m) or at least report CD for points beyond 100 m. The same issue applies to the L1 distance column.
  5. [Sec. 5.3, Tables 3 and 4] The claim of '30.5% Chamfer Distance improvement' is not clearly traceable to Table 3. Comparing the best LRS4Fusion result (1 s history, 3 s horizon, CD 39.031) to the best 4DOcc result (3 s history, 3 s horizon, CD 47.81) gives about 18%, not 30.5%. The 30.5% figure may correspond to a different comparison, but the table and text do not make this explicit. Please specify the exact comparison and compute the percentage consistently. Similar ambiguity appears in the NuScenes numbers in the text (e.g., '29%' which seems to correspond to the 3 s-in, 3 s-out row).
minor comments (5)
  1. [Sec. 4] Several numerical values are missing from the dataset description (e.g., '������ unlabeled frames', '������ manually annotated frames', '������ at ���'). The reader cannot assess the dataset scale. Please ensure all numbers render correctly.
  2. [Sec. 5.3, Table 4] The table layout is misaligned: for the 0 s history row, the LRS4Fusion values appear in different columns than HERMES, and the 'Horizon' column seems to contain both history and future horizon values. This makes the comparison difficult to interpret.
  3. [Abstract and throughout] Minor typos: 'an 26.6%' should be 'a 26.6%'. The conclusion states the method is 'a long-range camera-LiDAR BEV method' although the method is explicitly sparse voxel-based; consider rewording for consistency.
  4. [Sec. 3.1, Eq. (6)] The attention equation uses a softmax over '3D window of voxels' but the normalization factor is described as 'the dimensionality of the hidden features'; it is unclear whether this is a full softmax normalization or a scaled dot-product. Please clarify the notation and the role of the softmax normalization factor.
  5. [Sec. 5.1, Table 1] The depth ablation in Table 6 shows that using ResNet50 instead of Vim improves runtime and MAE slightly, but the paper does not discuss why Vim is nevertheless chosen for the final model beyond memory. A brief clarification would help.

Circularity Check

0 steps flagged · score 2.0 of 10

No derivation chain to be circular; self-citations are baselines/related work, not load-bearing; headline validation gap is not circularity.

full rationale

LRS4Fusion is an empirical systems paper; there is no analytic derivation whose conclusion is equivalent to its premises. The self-supervised pretraining predicts future LiDAR occupancy/velocity supervised directly by recorded LiDAR (Sec. 3.2, Fig. 5), and the downstream forecasting/OD tasks are evaluated against external ground truth (Sec. 5.2–5.3); the future LiDAR target is not an output used to define the model. The strongest numerical claims (52.61 mAP vs SAMFusion 41.55; CD 15.821) compare against published or reproducible baselines on a private dataset. The fact that SAMFusion is the authors' own prior work is a self-citation, but it is used as a baseline rather than as justification, and it is not the only baseline (PointPillars, BEVFormer, BEVFusion are also compared). The absence of range-resolved evaluation—aggregate mAP over a test set concentrated at close range (Fig. 2)—is a validation gap and a correctness risk, but it does not make any reported number equal to its input by construction. No fitted parameter is renamed as a prediction; no uniqueness theorem is imported; no ansatz is smuggled in via self-citation. Score 2 reflects one minor self-citation (SAMFusion) that is not load-bearing rather than any circular step.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

No physical entities are introduced. The free parameters are network design choices and evaluation regions chosen for the new dataset; they are not fitted constants in a derivation but they do affect the headline numbers.

free parameters (4)
  • History horizon = 1 s
    Selected by ablation on the private dataset (Table 7c); the claimed mAP gain from self-supervision depends on this choice.
  • Sparse voxel scale configuration = 4 scales, sizes not stated in main text
    The sparse representation uses 4 voxel scales and the temporal attention only on the second-smallest scale; these choices are tuned for memory and accuracy and appear without ablations.
  • Sparse attention window size = 3x3x3 in Figure 4 example
    The local attention neighborhood is a hand-chosen hyperparameter; no window-size ablation is reported.
  • Detection and forecasting ROI = X in [-100, +250] m, Y in [-100, +100] m
    The long-range ROI is chosen to match the claim; metric values are sensitive to this boundary.
assumptions (4)
  • domain assumption FMCW LiDAR per-voxel radial velocities are reliable up to 250m and sufficient to warp past voxels
    Invoked in Eq. 5 and the Temporal Sparse Fusion paragraph; degraded Doppler at long range would break temporal alignment.
  • domain assumption Ray-based free-space labels from LiDAR provide correct occupancy and velocity self-supervision
    Section 3.2: positions along each LiDAR ray are labeled unoccupied; occluded or low-reflectivity objects are mislabeled.
  • domain assumption Private dataset annotations of 7 classes up to 400m are accurate and complete
    Section 4; at 250-400m humans may miss small objects, inflating mAP.
  • ad hoc to paper Zero-padding camera or LiDAR-empty voxels before sparse convolution is a valid fusion strategy
    Section 3.1 Camera-LiDAR Fusion; this architectural choice is not compared to alternatives.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Self-Supervised Sparse Sensor Fusion for Long Range Perception." pith.science (2026). https://pith.science/paper/KNQUKGUX

@misc{pith2026250813995,
  author       = {Pith},
  title        = {Pith review of: Self-Supervised Sparse Sensor Fusion for Long Range Perception},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KNQUKGUX}},
  note         = {Machine review of arXiv:2508.13995}
}
read the original abstract

Outside of urban hubs, autonomous cars and trucks have to master driving on intercity highways. Safe, long-distance highway travel at speeds exceeding 100 km/h demands perception distances of at least 250 m, which is about five times the 50-100m typically addressed in city driving, to allow sufficient planning and braking margins. Increasing the perception ranges also allows to extend autonomy from light two-ton passenger vehicles to large-scale forty-ton trucks, which need a longer planning horizon due to their high inertia. However, most existing perception approaches focus on shorter ranges and rely on Bird's Eye View (BEV) representations, which incur quadratic increases in memory and compute costs as distance grows. To overcome this limitation, we built on top of a sparse representation and introduced an efficient 3D encoding of multi-modal and temporal features, along with a novel self-supervised pre-training scheme that enables large-scale learning from unlabeled camera-LiDAR data. Our approach extends perception distances to 250 meters and achieves an 26.6% improvement in mAP in object detection and a decrease of 30.5% in Chamfer Distance in LiDAR forecasting compared to existing methods, reaching distances up to 250 meters. Project Page: https://light.princeton.edu/lrs4fusion/

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

83 extracted references · 65 canonical work pages

  1. [1]

    Uno: Unsupervised occupancy fields for perception and forecasting

    Ben Agro, Quinlan Sykora, Sergio Casas, Thomas Gilles, and Raquel Urtasun. Uno: Unsupervised occupancy fields for perception and forecasting. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 14487–14496, 2024. 1, 2, 3, 5, 7

    2024

  2. [2]

    Depth pro: Sharp monocular metric depth in less than a second

    Aleksei Bochkovskii, Ama ¨el Delaunoy, Hugo Germain, Marcel Santos, Yichao Zhou, Stephan R Richter, and Vladlen Koltun. Depth pro: Sharp monocular metric depth in less than a second. arXiv preprint arXiv:2410.02073, 2024. 3

    arXiv 2024

  3. [3]

    Also: Automotive lidar self- supervision by occupancy estimation

    Alexandre Boulch, Corentin Sautier, Bj ¨orn Michele, Gilles Puy, and Renaud Marlet. Also: Automotive lidar self- supervision by occupancy estimation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 13455–13465, 2023. 3

    2023

  4. [4]

    Cross-spectral gated-rgb stereo depth estimation

    Samuel Brucker, Stefanie Walz, Mario Bijelic, and Felix Heide. Cross-spectral gated-rgb stereo depth estimation. In Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition, pages 21654–21665, 2024. 3

    2024

  5. [5]

    Lang, Sourabh V ora, Venice Erin Liong, Qiang Xu, Anush Krishnan, Yu Pan, Gi- ancarlo Baldan, and Oscar Beijbom

    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, 2020. 1, 2, 6, 7, 8

    work page 2020

  6. [6]

    Bevfusion4d: Learn- ing lidar-camera fusion under bird’s-eye-view via cross- modality guidance and temporal aggregation

    Hongxiang Cai, Zeyuan Zhang, Zhenyu Zhou, Ziyin Li, Wenbo Ding, and Jiuhua Zhao. Bevfusion4d: Learn- ing lidar-camera fusion under bird’s-eye-view via cross- modality guidance and temporal aggregation. arXiv preprint arXiv:2303.17099, 2023. 2

    arXiv 2023

  7. [7]

    Monoscene: Monoc- ular 3d semantic scene completion

    Anh-Quan Cao and Raoul De Charette. Monoscene: Monoc- ular 3d semantic scene completion. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 3991–4001, 2022. 2

    work page 2022

  8. [8]

    Ar- goverse: 3d tracking and forecasting with rich maps

    Ming-Fang Chang, John Lambert, Patsorn Sangkloy, Jag- jeet Singh, Slawomir Bak, Andrew Hartnett, De Wang, Peter Carr, Simon Lucey, Deva Ramanan, and James Hays. Ar- goverse: 3d tracking and forecasting with rich maps. In 2019 IEEE/CVF Conference on Computer Vision and Pat- tern Recognition (CVPR), pages 8740–8749, 2019. 6

    work page 2019

Show all 83 references
  1. [9]

    Improved baselines with momentum contrastive learning

    Xinlei Chen, Haoqi Fan, Ross Girshick, and Kaiming He. Improved baselines with momentum contrastive learning. arXiv preprint arXiv:2003.04297, 2020. 3

    2003 arXiv

  2. [10]

    Bert: Pre-training of deep bidirectional transformers for language understanding

    Jacob Devlin. Bert: Pre-training of deep bidirectional transformers for language understanding. arXiv preprint arXiv:1810.04805, 2018. 3

    2018 arXiv

  3. [11]

    Are we ready for autonomous driving? the kitti vision benchmark suite

    Andreas Geiger, Philip Lenz, and Raquel Urtasun. Are we ready for autonomous driving? the kitti vision benchmark suite. In 2012 IEEE conference on computer vision and pat- tern recognition, pages 3354–3361. IEEE, 2012. 6

    2012

  4. [12]

    Vip3d: End-to-end visual trajectory prediction via 3d agent queries

    Junru Gu, Chenxu Hu, Tianyuan Zhang, Xuanyao Chen, Yilun Wang, Yue Wang, and Hang Zhao. Vip3d: End-to-end visual trajectory prediction via 3d agent queries. InProceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 5496–5506, 2023. 2

    2023

  5. [13]

    Momentum contrast for unsupervised visual rep- resentation learning

    Kaiming He, Haoqi Fan, Yuxin Wu, Saining Xie, and Ross Girshick. Momentum contrast for unsupervised visual rep- resentation learning. In Proceedings of the IEEE/CVF con- ference on computer vision and pattern recognition , pages 9729–9738, 2020. 3

    2020

  6. [14]

    Masked autoencoders are scalable vision learners

    Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Doll´ar, and Ross Girshick. Masked autoencoders are scalable vision learners. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages 16000– 16009, 2022. 3

    2022

  7. [15]

    Masked autoencoder for self-supervised pre-training on lidar point clouds

    Georg Hess, Johan Jaxing, Elias Svensson, David Hager- man, Christoffer Petersson, and Lennart Svensson. Masked autoencoder for self-supervised pre-training on lidar point clouds. In Proceedings of the IEEE/CVF winter conference on applications of computer vision, pages 350–35...

    2023

  8. [16]

    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, 2

    2023

  9. [17]

    Tri-perspective view for vision- based 3d semantic occupancy prediction

    Yuanhui Huang, Wenzhao Zheng, Yunpeng Zhang, Jie Zhou, and Jiwen Lu. Tri-perspective view for vision- based 3d semantic occupancy prediction. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 9223–9232, 2023. 2

    2023

  10. [18]

    Selfocc: Self-supervised vision-based 3d oc- cupancy prediction

    Yuanhui Huang, Wenzhao Zheng, Borui Zhang, Jie Zhou, and Jiwen Lu. Selfocc: Self-supervised vision-based 3d oc- cupancy prediction. In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 19946–19956, 2024. 2

    2024

  11. [19]

    Gaussianformer: Scene as gaussians for vision-based 3d semantic occupancy prediction

    Yuanhui Huang, Wenzhao Zheng, Yunpeng Zhang, Jie Zhou, and Jiwen Lu. Gaussianformer: Scene as gaussians for vision-based 3d semantic occupancy prediction. arXiv preprint arXiv:2405.17429, 2024. 2

    2024 arXiv

  12. [20]

    Driveadapter: Breaking the coupling barrier of perception and planning in end-to-end autonomous driving

    Xiaosong Jia, Yulu Gao, Li Chen, Junchi Yan, Patrick Langechuan Liu, and Hongyang Li. Driveadapter: Breaking the coupling barrier of perception and planning in end-to-end autonomous driving. In Proceedings of the IEEE/CVF International Conference on Computer Vision , pages 795...

    2023

  13. [21]

    Think twice be- fore driving: Towards scalable decoders for end-to-end au- tonomous driving

    Xiaosong Jia, Penghao Wu, Li Chen, Jiangwei Xie, Con- ghui He, Junchi Yan, and Hongyang Li. Think twice be- fore driving: Towards scalable decoders for end-to-end au- tonomous driving. In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, page...

    2023

  14. [22]

    Vad: Vectorized scene representa- tion for efficient autonomous driving

    Bo Jiang, Shaoyu Chen, Qing Xu, Bencheng Liao, Jiajie Chen, Helong Zhou, Qian Zhang, Wenyu Liu, Chang Huang, and Xinggang Wang. Vad: Vectorized scene representa- tion for efficient autonomous driving. In Proceedings of the IEEE/CVF International Conference on Computer Vision ,...

    2023

  15. [23]

    Supervised contrastive learning

    Prannay Khosla, Piotr Teterwak, Chen Wang, Aaron Sarna, Yonglong Tian, Phillip Isola, Aaron Maschinot, Ce Liu, and Dilip Krishnan. Supervised contrastive learning. Advances in neural information processing systems, 33:18661–18673,

  16. [24]

    Point cloud forecasting as a proxy for 4d occupancy forecasting, 2023

    Tarasha Khurana, Peiyun Hu, David Held, and Deva Ra- manan. Point cloud forecasting as a proxy for 4d occupancy forecasting, 2023. 7

    2023

  17. [25]

    In defense of classical image processing: Fast depth completion on the cpu

    Jason Ku, Ali Harakeh, and Steven L Waslander. In defense of classical image processing: Fast depth completion on the cpu. In 2018 15th Conference on Computer and Robot Vision (CRV), pages 16–22. IEEE, 2018. 3

    2018

  18. [26]

    Pointpillars: Fast encoders for object detection from point clouds

    Alex H Lang, Sourabh V ora, Holger Caesar, Lubing Zhou, Jiong Yang, and Oscar Beijbom. Pointpillars: Fast encoders for object detection from point clouds. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 12697–12705, 2019. 2, 6

    2019

  19. [27]

    Hdmapnet: An online HD map construction and evaluation framework

    Qi Li, Yue Wang, Yilun Wang, and Hang Zhao. Hdmapnet: An online HD map construction and evaluation framework. CoRR, abs/2107.06307, 2021. 2

    2021 arXiv

  20. [28]

    Unifying voxel-based representation with transformer for 3d object detection

    Yanwei Li, Yilun Chen, Xiaojuan Qi, Zeming Li, Jian Sun, and Jiaya Jia. Unifying voxel-based representation with transformer for 3d object detection. Advances in Neural In- formation Processing Systems, 35:18442–18455, 2022. 2

    2022

  21. [29]

    Fully sparse fusion for 3d object detection

    Yingyan Li, Lue Fan, Yang Liu, Zehao Huang, Yuntao Chen, Naiyan Wang, and Zhaoxiang Zhang. Fully sparse fusion for 3d object detection. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2024. 2

    2024

  22. [30]

    Bevformer: Learning bird’s-eye-view representation from multi-camera images via spatiotemporal transformers

    Zhiqi Li, Wenhai Wang, Hongyang Li, Enze Xie, Chong- hao Sima, Tong Lu, Yu Qiao, and Jifeng Dai. Bevformer: Learning bird’s-eye-view representation from multi-camera images via spatiotemporal transformers. In European con- ference on computer vision, pages 1–18. Springer, 2022...

    2022

  23. [31]

    Bevfusion: A simple and robust lidar-camera fusion framework

    Tingting Liang, Hongwei Xie, Kaicheng Yu, Zhongyu Xia, Zhiwei Lin, Yongtao Wang, Tao Tang, Bing Wang, and Zhi Tang. Bevfusion: A simple and robust lidar-camera fusion framework. Advances in Neural Information Processing Sys- tems, 35:10421–10434, 2022. 2

    2022

  24. [32]

    Feature pyramid networks for object detection

    Tsung-Yi Lin, Piotr Doll ´ar, Ross Girshick, Kaiming He, Bharath Hariharan, and Serge Belongie. Feature pyramid networks for object detection. In 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 936–944, 2017. 3

    2017

  25. [33]

    Raft-stereo: Multilevel recurrent field transforms for stereo matching

    Lahav Lipson, Zachary Teed, and Jia Deng. Raft-stereo: Multilevel recurrent field transforms for stereo matching. In 2021 International Conference on 3D Vision (3DV) , pages 218–227. IEEE, 2021. 3

    2021

  26. [34]

    Sparsebev: High-performance sparse 3d object de- tection from multi-camera videos

    Haisong Liu, Yao Teng, Tao Lu, Haiguang Wang, and Limin Wang. Sparsebev: High-performance sparse 3d object de- tection from multi-camera videos. In Proceedings of the IEEE/CVF International Conference on Computer Vision , pages 18580–18590, 2023. 2

    2023

  27. [35]

    Petrv2: A unified framework for 3d perception from multi-camera images

    Yingfei Liu, Junjie Yan, Fan Jia, Shuailin Li, Aqi Gao, Tian- cai Wang, and Xiangyu Zhang. Petrv2: A unified framework for 3d perception from multi-camera images. InProceedings of the IEEE/CVF International Conference on Computer Vi- sion, pages 3262–3272, 2023. 2

    2023

  28. [36]

    Bevfusion: Multi-task multi-sensor fusion with unified bird’s-eye view representation

    Zhijian Liu, Haotian Tang, Alexander Amini, Xinyu Yang, Huizi Mao, Daniela L Rus, and Song Han. Bevfusion: Multi-task multi-sensor fusion with unified bird’s-eye view representation. In 2023 IEEE International Conference on Robotics and Automation (ICRA) , pages 2774–2781. IEEE,

    2023

  29. [37]

    One million scenes for autonomous driving: Once dataset,

    Jiageng Mao, Minzhe Niu, Chenhan Jiang, Hanxue Liang, Jingheng Chen, Xiaodan Liang, Yamin Li, Chaoqiang Ye, Wei Zhang, Zhenguo Li, Jie Yu, Hang Xu, and Chunjing Xu. One million scenes for autonomous driving: Once dataset,

  30. [38]

    Dinov2: Learning robust visual features without supervision

    Maxime Oquab, Timoth ´ee Darcet, Th ´eo Moutakanni, Huy V o, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, et al. Dinov2: Learning robust visual features without supervision. arXiv preprint arXiv:2304.07193, 2023. 3

    2023 arXiv

  31. [39]

    Samfusion: Sensor-adaptive multi- modal fusion for 3d object detection in adverse weather

    Edoardo Palladin, Roland Dietze, Praveen Narayanan, Mario Bijelic, and Felix Heide. Samfusion: Sensor-adaptive multi- modal fusion for 3d object detection in adverse weather. In Proceedings of the IEEE European Conference on Computer Vision (ECCV), 2024. 1, 2, 3, 6, 7

    2024

  32. [40]

    Non-local spatial propagation network for depth completion

    Jinsun Park, Kyungdon Joo, Zhe Hu, Chi-Kuei Liu, and In So Kweon. Non-local spatial propagation network for depth completion. In Computer Vision–ECCV 2020: 16th Euro- pean Conference, Glasgow, UK, August 23–28, 2020, Pro- ceedings, Part XIII 16, pages 120–136. Springer, 2020. 3

    2020

  33. [41]

    Bevsegformer: Bird’s eye view semantic segmentation from arbitrary camera rigs, 2022

    Lang Peng, Zhirong Chen, Zhangjie Fu, Pengpeng Liang, and Erkang Cheng. Bevsegformer: Bird’s eye view semantic segmentation from arbitrary camera rigs, 2022. 2

    2022

  34. [42]

    Lift, splat, shoot: Encod- ing images from arbitrary camera rigs by implicitly unpro- jecting to 3d

    Jonah Philion and Sanja Fidler. Lift, splat, shoot: Encod- ing images from arbitrary camera rigs by implicitly unpro- jecting to 3d. In Computer Vision–ECCV 2020: 16th Euro- pean Conference, Glasgow, UK, August 23–28, 2020, Pro- ceedings, Part XIV 16 , pages 194–210. Springer,...

    2020

  35. [43]

    Unidepth: Universal monocular metric depth estimation

    Luigi Piccinelli, Yung-Hsu Yang, Christos Sakaridis, Mattia Segu, Siyuan Li, Luc Van Gool, and Fisher Yu. Unidepth: Universal monocular metric depth estimation. In Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 10106–10116, 2024. 3

    2024

  36. [44]

    Pointnet: Deep learning on point sets for 3d classification and segmentation

    Charles R Qi, Hao Su, Kaichun Mo, and Leonidas J Guibas. Pointnet: Deep learning on point sets for 3d classification and segmentation. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 652–660,

  37. [45]

    From points to parts: 3d object detec- tion from point cloud with part-aware and part-aggregation network, 2020

    Shaoshuai Shi, Zhe Wang, Jianping Shi, Xiaogang Wang, and Hongsheng Li. From points to parts: 3d object detec- tion from point cloud with part-aware and part-aggregation network, 2020. 4

    2020

  38. [46]

    Scalability in perception for au- tonomous driving: Waymo open dataset, 2020

    Pei Sun, Henrik Kretzschmar, Xerxes Dotiwalla, Aurelien Chouard, Vijaysai Patnaik, Paul Tsui, James Guo, Yin Zhou, Yuning Chai, Benjamin Caine, Vijay Vasudevan, Wei Han, Jiquan Ngiam, Hang Zhao, Aleksei Timofeev, Scott Ettinger, Maxim Krivokon, Amy Gao, Aditya Joshi, Sheng Zha...

    2020

  39. [47]

    Learning guided convolutional network for depth comple- tion

    Jie Tang, Fei-Peng Tian, Wei Feng, Jian Li, and Ping Tan. Learning guided convolutional network for depth comple- tion. IEEE Transactions on Image Processing , 30:1116– 1129, 2020. 3

    2020

  40. [48]

    Sparseocc: Re- thinking sparse latent representation for vision-based seman- tic occupancy prediction

    Pin Tang, Zhongdao Wang, Guoqing Wang, Jilai Zheng, Xi- angxuan Ren, Bailan Feng, and Chao Ma. Sparseocc: Re- thinking sparse latent representation for vision-based seman- tic occupancy prediction. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogn...

    2024

  41. [49]

    Raft: Recurrent all-pairs field transforms for optical flow

    Zachary Teed and Jia Deng. Raft: Recurrent all-pairs field transforms for optical flow. In Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23– 28, 2020, Proceedings, Part II 16, pages 402–419. Springer,

    2020

  42. [50]

    Vision-based intention and trajectory prediction in autonomous vehicles: A survey

    Izzeddin Teeti, Salman Khan, Ajmal Shahbaz, Andrew Bradley, Fabio Cuzzolin, and Lud De Raedt. Vision-based intention and trajectory prediction in autonomous vehicles: A survey. In IJCAI, pages 5630–5637, 2022. 2

    2022

  43. [51]

    Con- trastive multiview coding

    Yonglong Tian, Dilip Krishnan, and Phillip Isola. Con- trastive multiview coding. In Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part XI 16 , pages 776–794. Springer,

    2020

  44. [52]

    Scene as occupancy

    Wenwen Tong, Chonghao Sima, Tai Wang, Li Chen, Silei Wu, Hanming Deng, Yi Gu, Lewei Lu, Ping Luo, Dahua Lin, et al. Scene as occupancy. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 8406– 8415, 2023. 1

    2023

  45. [53]

    Pointpainting: Sequential fusion for 3d object de- tection

    Sourabh V ora, Alex H Lang, Bassam Helou, and Oscar Bei- jbom. Pointpainting: Sequential fusion for 3d object de- tection. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 4604–4612,

  46. [54]

    Gated stereo: Joint depth estimation from gated and wide-baseline active stereo cues

    Stefanie Walz, Mario Bijelic, Andrea Ramazzina, Aman- preet Walia, Fahim Mannan, and Felix Heide. Gated stereo: Joint depth estimation from gated and wide-baseline active stereo cues. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 1...

    2023

  47. [55]

    Pointaugmenting: Cross-modal augmentation for 3d object detection

    Chunwei Wang, Chao Ma, Ming Zhu, and Xiaokang Yang. Pointaugmenting: Cross-modal augmentation for 3d object detection. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 11794– 11803, 2021. 2

    2021

  48. [56]

    Waslander, Yue Wang, Sanja Fi- dler, Marco Pavone, and Peter Karkus

    Letian Wang, Seung Wook Kim, Jiawei Yang, Cunjun Yu, Boris Ivanovic, Steven L. Waslander, Yue Wang, Sanja Fi- dler, Marco Pavone, and Peter Karkus. Distillnerf: Perceiv- ing 3d scenes from single-glance images by distilling neural fields and foundation model features, 2024. 2, 3

    2024

  49. [57]

    Openoccupancy: A large scale benchmark for surrounding semantic occupancy perception

    Xiaofeng Wang, Zheng Zhu, Wenbo Xu, Yunpeng Zhang, Yi Wei, Xu Chi, Yun Ye, Dalong Du, Jiwen Lu, and Xin- gang Wang. Openoccupancy: A large scale benchmark for surrounding semantic occupancy perception. In Proceedings of the IEEE/CVF International Conference on Computer Vi- sio...

    2023

  50. [58]

    Masked feature predic- tion for self-supervised visual pre-training

    Chen Wei, Haoqi Fan, Saining Xie, Chao-Yuan Wu, Alan Yuille, and Christoph Feichtenhofer. Masked feature predic- tion for self-supervised visual pre-training. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 14668–14678, 2022. 3

    2022

  51. [59]

    Surroundocc: Multi-camera 3d occu- pancy prediction for autonomous driving

    Yi Wei, Linqing Zhao, Wenzhao Zheng, Zheng Zhu, Jie Zhou, and Jiwen Lu. Surroundocc: Multi-camera 3d occu- pancy prediction for autonomous driving. In Proceedings of the IEEE/CVF International Conference on Computer Vi- sion, pages 21729–21740, 2023. 2

    2023

  52. [60]

    Inverting the pose forecasting pipeline with spf2: Sequential pointcloud forecasting for se- quential pose forecasting, 2020

    Xinshuo Weng, Jianren Wang, Sergey Levine, Kris Kitani, and Nicholas Rhinehart. Inverting the pose forecasting pipeline with spf2: Sequential pointcloud forecasting for se- quential pose forecasting, 2020. 7

    2020

  53. [61]

    Xinshuo Weng, Junyu Nan, Kuan-Hui Lee, Rowan McAllis- ter, Adrien Gaidon, Nicholas Rhinehart, and Kris M. Kitani. S2net: Stochastic sequential pointcloud forecasting. In Com- puter Vision – ECCV 2022: 17th European Conference, Tel Aviv, Israel, October 23–27, 2022, Proceedings...

    2022

  54. [62]

    Para-drive: Parallelized architecture for real- time autonomous driving

    Xinshuo Weng, Boris Ivanovic, Yan Wang, Yue Wang, and Marco Pavone. Para-drive: Parallelized architecture for real- time autonomous driving. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 15449–15458, 2024. 1, 2

    2024

  55. [63]

    Argoverse 2: Next generation datasets for self-driving perception and fore- casting, 2023

    Benjamin Wilson, William Qi, Tanmay Agarwal, John Lam- bert, Jagjeet Singh, Siddhesh Khandelwal, Bowen Pan, Rat- nesh Kumar, Andrew Hartnett, Jhony Kaesemodel Pontes, Deva Ramanan, Peter Carr, and James Hays. Argoverse 2: Next generation datasets for self-driving perception an...

    2023

  56. [64]

    Simmim: A simple framework for masked image modeling

    Zhenda Xie, Zheng Zhang, Yue Cao, Yutong Lin, Jianmin Bao, Zhuliang Yao, Qi Dai, and Han Hu. Simmim: A simple framework for masked image modeling. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 9653–9663, 2022. 3

    2022

  57. [65]

    Iterative geometry encoding volume for stereo matching

    Gangwei Xu, Xianqi Wang, Xiaohuan Ding, and Xin Yang. Iterative geometry encoding volume for stereo matching. In Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition, pages 21919–21928, 2023. 3

    2023

  58. [66]

    Fusionpainting: Multimodal fusion with adaptive attention for 3d object detection

    Shaoqing Xu, Dingfu Zhou, Jin Fang, Junbo Yin, Zhou Bin, and Liangjun Zhang. Fusionpainting: Multimodal fusion with adaptive attention for 3d object detection. In2021 IEEE International Intelligent Transportation Systems Conference (ITSC), pages 3047–3054. IEEE, 2021. 2

    2021

  59. [67]

    Unipad: A universal pre-training paradigm for autonomous driving

    Honghui Yang, Sha Zhang, Di Huang, Xiaoyang Wu, Haoyi Zhu, Tong He, Shixiang Tang, Hengshuang Zhao, Qibo Qiu, Binbin Lin, et al. Unipad: A universal pre-training paradigm for autonomous driving. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogniti...

    2024

  60. [68]

    Depth anything: Unleashing the power of large-scale unlabeled data

    Lihe Yang, Bingyi Kang, Zilong Huang, Xiaogang Xu, Jiashi Feng, and Hengshuang Zhao. Depth anything: Unleashing the power of large-scale unlabeled data. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 10371–10381, 2024. 3

    2024

  61. [69]

    Deepinteraction: 3d object detection via modality interaction

    Zeyu Yang, Jiaqi Chen, Zhenwei Miao, Wei Li, Xiatian Zhu, and Li Zhang. Deepinteraction: 3d object detection via modality interaction. Advances in Neural Information Pro- cessing Systems, 35:1992–2005, 2022. 1

    1992

  62. [70]

    Visual point cloud forecasting enables scalable autonomous driving

    Zetong Yang, Li Chen, Yanan Sun, and Hongyang Li. Visual point cloud forecasting enables scalable autonomous driving. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 14673–14684, 2024. 1, 2, 3, 6, 7

    2024

  63. [71]

    Is-fusion: Instance-scene collaborative fusion for multimodal 3d ob- ject detection

    Junbo Yin, Jianbing Shen, Runnan Chen, Wei Li, Ruigang Yang, Pascal Frossard, and Wenguan Wang. Is-fusion: Instance-scene collaborative fusion for multimodal 3d ob- ject detection. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 1490...

    2024

  64. [72]

    Multi- modal virtual point 3d detection

    Tianwei Yin, Xingyi Zhou, and Philipp Kr ¨ahenb¨uhl. Multi- modal virtual point 3d detection. Advances in Neural Infor- mation Processing Systems, 34:16494–16507, 2021. 2

    2021

  65. [73]

    Center- based 3d object detection and tracking, 2021

    Tianwei Yin, Xingyi Zhou, and Philipp Kr ¨ahenb¨uhl. Center- based 3d object detection and tracking, 2021. 6

    2021

  66. [74]

    Metric3d: Towards zero-shot metric 3d prediction from a single image

    Wei Yin, Chi Zhang, Hao Chen, Zhipeng Cai, Gang Yu, Kaixuan Wang, Xiaozhi Chen, and Chunhua Shen. Metric3d: Towards zero-shot metric 3d prediction from a single image. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 9043–9053, 2023. 3

    2023

  67. [75]

    3d-cvf: Generating joint camera and lidar features us- ing cross-view spatial feature fusion for 3d object detection

    Jin Hyeok Yoo, Yecheol Kim, Jisong Kim, and Jun Won Choi. 3d-cvf: Generating joint camera and lidar features us- ing cross-view spatial feature fusion for 3d object detection. In Computer Vision–ECCV 2020: 16th European Confer- ence, Glasgow, UK, August 23–28, 2020, Proceeding...

    2020

  68. [76]

    Mutr3d: A multi-camera tracking frame- work via 3d-to-2d queries

    Tianyuan Zhang, Xuanyao Chen, Yue Wang, Yilun Wang, and Hang Zhao. Mutr3d: A multi-camera tracking frame- work via 3d-to-2d queries. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 4537–4546, 2022. 1

    2022

  69. [77]

    Completionformer: Depth completion with convolutions and vision transform- ers

    Youmin Zhang, Xianda Guo, Matteo Poggi, Zheng Zhu, Guan Huang, and Stefano Mattoccia. Completionformer: Depth completion with convolutions and vision transform- ers. In Proceedings of the IEEE/CVF conference on com- puter vision and pattern recognition , pages 18527–18536,

  70. [78]

    Minimal gated unit for recurrent neural networks

    Guo-Bing Zhou, Jianxin Wu, Chen-Lin Zhang, and Zhi-Hua Zhou. Minimal gated unit for recurrent neural networks. In- ternational Journal of Automation and Computing , 13(3): 226–234, 2016. 3

    2016

  71. [79]

    Hermes: A unified self-driving world model for simultaneous 3d scene understanding and generation, 2025

    Xin Zhou, Dingkang Liang, Sifan Tu, Xiwu Chen, Yikang Ding, Dingyuan Zhang, Feiyang Tan, Hengshuang Zhao, and Xiang Bai. Hermes: A unified self-driving world model for simultaneous 3d scene understanding and generation, 2025. 7

    2025

  72. [80]

    V oxelnet: End-to-end learning for point cloud based 3d object detection

    Yin Zhou and Oncel Tuzel. V oxelnet: End-to-end learning for point cloud based 3d object detection. In Proceedings of the IEEE conference on computer vision and pattern recog- nition, pages 4490–4499, 2018. 2

    2018

  73. [81]

    Vision mamba: Efficient visual representation learning with bidirectional state space model

    Lianghui Zhu, Bencheng Liao, Qian Zhang, Xinlong Wang, Wenyu Liu, and Xinggang Wang. Vision mamba: Efficient visual representation learning with bidirectional state space model. arXiv preprint arXiv:2401.09417, 2024. 3, 7, 8

    2024 arXiv

  74. [82]

    Deformable detr: Deformable trans- formers for end-to-end object detection

    Xizhou Zhu, Weijie Su, Lewei Lu, Bin Li, Xiaogang Wang, and Jifeng Dai. Deformable detr: Deformable trans- formers for end-to-end object detection. arXiv preprint arXiv:2010.04159, 2020. 2

    2010 arXiv

  75. [83]

    Ogni-dc: Robust depth com- pletion with optimization-guided neural iterations

    Yiming Zuo and Jia Deng. Ogni-dc: Robust depth com- pletion with optimization-guided neural iterations. Proceed- ings of the IEEE European Conference on Computer Vision (ECCV), 2024. 3, 6, 7, 8

    2024

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.