Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

Rethinking Temporal Fusion with a Unified Gradient Descent View for 3D Semantic Occupancy Prediction

T0 review · 3 major / 5 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read Reinterpreting temporal fusion as gradient descent lifts 3D occupancy mIoU by up to 4.8% while cutting memory by 27–72%.

desk verdict Solid empirical paper dressed in an oversold theory; the fusion modules work and the memory savings are real, but baseline fairness and code release need to be settled before the numbers are taken at face value. read the letter →

arxiv 2504.12959 v2 pith:POGCPCPS submitted 2025-04-17 cs.CV

Dubing Chen , Huan Zheng , Jin Fang , Xingping Dong , Xianfei Li , Wenlong Liao , Tao He , Pai Peng
show 1 more author
Jianbing Shen
This is my paper · ORCID
classification cs.CV
keywords 3Dsemanticoccupancypredictiontemporalfusionrecurrentneuralnetworksgradientdescentsceneadaptationmotioncalibrationgeometricpriorsautonomousdriving
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

The paper claims that temporal fusion for vision-based 3D semantic occupancy prediction (predicting the occupied voxels and their classes around a car from cameras) is currently too narrow: existing methods merge voxel features across frames but ignore three further kinds of history. It identifies scene-level consistency (weather, lighting, road conditions that persist over seconds), motion calibration (historical flow that corrects frame-to-frame alignment errors), and geometric complementation (past depth priors that improve the 2D-to-3D lifting step). To fuse these heterogeneous representations, it reinterprets the vanilla RNN update as a single gradient-descent step on a squared-error loss between current and historical features, then builds one loss per cue whose gradient acts as the fusion residual. If correct, the method raises mean IoU by 1.4–4.8 points over five reimplemented baselines on Occ3D and by 6.3–14.0 points over non-temporal baselines on SurroundOcc and OpenOccupancy, while cutting inference memory by 27–72% because all history is compressed into single-frame-sized hidden states rather than stored frame stacks. The payoff for autonomous driving is more accurate 3D scene understanding at lower memory cost, achieved by a plug-in module that works with existing occupancy pipelines.

What carries the argument

The load-bearing identity is Proposition 1: an RNN step $h_t = A h_{t-1} + B x_t$ is equivalent to one gradient-descent update of $h_{t-1}$ on the loss $\|A h_{t-1} - B x_t\|^2$. This converts 'how do we fuse heterogeneous temporal signals?' into 'which loss do we minimize?', so every cue gets a purpose-built loss whose gradient is the temporal residual added to the representation. Concretely: scene-level fusion minimizes a self-supervised augmentation-consistency loss over learnable layer-norm parameters $(\gamma, \beta, W, b)$; motion fusion minimizes $\|\mathrm{Warp}(H_m^{t-1}) - M_t\|^2$ with the gradient computed through the sampling Jacobian; geometry fusion blends warped historical depth histograms with the current estimate using a sigmoid-gated adaptive weight; and voxel fusion keeps a single-frame recurrent state $H_v^t = A_v\,\mathrm{warp}(H_v^{t-1}) + B_v V^t$ in place of a multi-frame queue. The single-frame state is what delivers the memory reduction, and the gradient-as-residual view is what lets all four cues share one RNN-style update rule.

What would settle it

Retrain the strongest baseline (ALOcc) under its own original published schedule until converged, add GDFusion, and compare mIoU and memory on Occ3D: if the gain over the properly trained baseline collapses, the reported improvement was partly a training-protocol artifact. A complementary check: run FB-Occ-GF on SurroundOcc, which the paper omits because of an implementation bug; the claim that gains are consistent across baselines and benchmarks requires this pairing to improve mIoU as well.

Watch

Extended reading notes

Core claim

The paper's central claim is that temporal fusion for VisionOcc should happen at four levels, not one: voxel-level volume features, scene-level network parameters that adapt to stable environmental conditions, motion fields that correct coordinate alignment, and probabilistic geometry priors that strengthen 2D-to-3D lifting. The unifying device is the observation that a vanilla RNN update $h_t = A h_{t-1} + B x_t$ is exactly one gradient-descent step on $h_{t-1}$ minimizing $\|A h_{t-1} - B x_t\|^2$ (Proposition 1). Each temporal cue is then fused by defining an MSE-style discrepancy between the current representation and the warped historical state, computing its gradient, and using that gradient as the update residual: scene parameters are adapted by a self-supervised augmentation-consistency loss, motion by an alignment loss whose gradient flows through the trilinear-sampling Jacobian, geometry by EMA-style blending with a sigmoid-gated adaptive rate, and voxels by a recurrent state of single-frame size. The paper demonstrates the scheme by wrapping it around five existing occupancy baselines and reporting consistent mIoU gains on all three nuScenes-based benchmarks, with memory savings of 27–72% on Occ3D because the multi-frame history queue is replaced by one frame-sized hidden state.

Load-bearing premise

The headline accuracy numbers compare GDFusion against baselines reimplemented under one uniform 12-epoch training schedule; if that schedule trains a baseline worse than its own published recipe, part of the measured mIoU gain could reflect the training protocol rather than the fusion module itself.

Editorial extensions

If this is right

  • Temporal fusion need not store $N_h$ historical volume frames: a single-frame-sized RNN hidden state compresses the whole history, so longer effective memory costs almost nothing extra (27–72% less inference memory on Occ3D).
  • The three new cues are complementary and additive: ablations show removing any one of scene, motion, or geometry fusion lowers accuracy, while the full combination reaches 43.3 mIoU relative to 38.0 for the compressed baseline without fusion.
  • GDFusion generalizes as a plug-in: it improves mIoU for all five reimplemented baselines on Occ3D (from +0.6 for ALOcc-mini to +1.9 for FB-Occ) and lifts non-temporal baselines by 6.3–14.0 points on SurroundOcc and OpenOccupancy with negligible added inference time.
  • Motion fusion needs no flow supervision: the learned offsets behave like deformable-convolution sampling corrections, so historical motion improves alignment without extra labels.
  • The scene-adaptation module keeps updating parameters at inference time, meaning the network continuously adapts to the current scene rather than freezing after training.

Reading between the lines

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

  • The Proposition 1 recipe — define an MSE loss between current and warped history, use its gradient as the residual — is representation-agnostic, so the same scheme could fuse heterogeneous temporal states in other streaming perception tasks such as BEV detection, HD mapping, or panoptic segmentation, not just occupancy.
  • The scene-level module is in effect a running test-time adaptation; a direct domain-shift experiment (e.g., night or rain sequences) would test whether the claimed scene-consistency cue measurably improves robustness, something the paper motivates but does not measure.
  • The quoted 27–72% memory savings are measured at specific voxel resolutions on Occ3D; a fuller picture would report how the saving scales as occupancy resolution, sequence length, or number of cameras grows.
  • The paper's appendix shows the plain vanilla recurrence beating modern state-space models (Mamba, xLSTM, RWKV) on nuScenes' short sequences; on longer-horizon driving logs, those expressive recurrent states might overtake the simple gradient-descent form — a natural follow-up test.
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 / 5 minor

Summary. The paper proposes GDFusion, a temporal fusion framework for vision-based 3D semantic occupancy prediction (VisionOcc). It identifies three temporal cues beyond standard voxel-level fusion: scene-level consistency, motion calibration, and geometric complementation. These cues are encoded in heterogeneous representations (scene-adaptive network parameters, motion maps, geometry distributions) and integrated through RNN-style updates that the authors reinterpret as gradient descent steps on discrepancy losses. The method is evaluated on Occ3D, SurroundOcc, and OpenOccupancy by plugging it into five reimplemented baselines (BEVDetOcc, FB-Occ, ALOcc, ALOcc-mini, ALOcc-2D). The central claims are consistent mIoU improvements over the baselines and a substantial reduction in inference memory (27%-72% on Occ3D) compared with multi-frame stacking approaches such as SOLOFusion.

Significance. If the empirical results hold, GDFusion is a practically useful plug-and-play temporal fusion module for VisionOcc: it reports gains across three benchmarks, detailed ablations, per-class results, RayIoU evaluations, and a modern-RNN comparison, and the memory-efficiency improvement is large and clearly demonstrated. The exploration of scene-level, motion, and geometry cues as complementary temporal signals is a valuable contribution that goes beyond voxel feature stacking. However, the theoretical framing in Proposition 1 is weaker than claimed, and the central empirical comparison depends on the fairness of the uniform reimplementation protocol, which is not yet fully established. The paper does not ship code or official checkpoints, so the reimplementation details are important for verification.

major comments (3)
  1. [Section 5, Implementation Details and Tables 1-3] The central empirical claim is conditioned on the fairness of the uniform reimplementation protocol. All five baselines and their GDFusion variants were trained for 12 epochs with CBGS, learning rate 2e-4, batch size 16, and video-level shuffling, while FB-Occ and ALOcc were originally published with their own training schedules. The paper provides no evidence that 12 epochs is sufficient for each baseline under this protocol, and it does not report comparisons against the official checkpoints of FB-Occ or ALOcc. As a result, the reported mIoU gains (e.g., +1.9 for FB-Occ-GF in Table 1) could be inflated by undertrained baselines rather than by the fusion modules themselves. In addition, FB-Occ is omitted on SurroundOcc because of an "unexpected bug in its implementation" (Section 5.2), which removes one of the claimed cross-benchmark confirmations. Please add official-checkpoint comparisons, per-baseline training curves with the uniform protocol, or experiments with the original schedules to rule out protocol effects.
  2. [Section 4.1, Proposition 1] The claimed equivalence between an RNN update and a gradient descent step is only established in one direction, and that direction restricts the class of representable RNN matrices. The proof shows that a gradient descent step on ||A h_{t-1} - B x_t||^2 produces h_t = (I - 2 eta A^T A) h_{t-1} + 2 eta A^T B x_t, where the coefficient of h_{t-1} is symmetric and I minus that matrix is positive semidefinite. An arbitrary RNN matrix A' need not satisfy this constraint, so the statement "The RNN update step h_t = A h_{t-1} + B x_t is equivalent to a gradient descent step" is not correct as stated. This does not invalidate the fusion update equations, which are self-contained, but it weakens the unified optimization interpretation that is presented as a central contribution. The proposition should be restated as a one-way construction, with a discussion of the representational restriction.
  3. [Section 4.3, Eq. (16) and Appendix A.2] The motion-Jacobian formula in Eq. (16) is not fully justified in the main text. The appendix derives the Jacobian J of the sampling function with respect to sampling coordinates in Eq. (A.49), but the chain rule through the coordinate map R_{t->t-1}(P + M_t) is not shown, and the appearance of R^T_{t->t-1} J - I with that particular ordering and transpose is non-obvious. If the formula is incorrect, the motion fusion update is not the claimed gradient descent step, and the ablation results for the motion cue would need reinterpretation. Please provide the complete vectorized derivation or correct the formula, and state the dimensions of each term.
minor comments (5)
  1. [Abstract and Section 5.2] The phrase "1.4%-4.8% mIoU improvements" is ambiguous: Table 1 reports absolute mIoU gains of 0.6-1.9 points, so the percentages appear to be relative improvements. Please label them as relative gains to avoid confusion.
  2. [Section 4.4, Eqs. (20)-(21)] The factor of 2 in the gradient is absorbed into eta_g without renaming the learning rate; state this explicitly so that Eq. (20) and Eq. (21) are not read as inconsistent.
  3. [Section 5.2] The omission of FB-Occ on SurroundOcc due to an "unexpected bug in its implementation" should be explained in enough detail for reproducibility, because the omission removes one of the cross-benchmark confirmations of the method's generality.
  4. [Tables 2 and 3] FPS and memory values in Tables 2 and 3 come from different hardware (A100, RTX 4090, V100) and different sources; the caption notes this, but the main text should explicitly warn against cross-table comparisons of these efficiency numbers.
  5. [Section 6] The conclusion refers to the method as "GradFusion" while the rest of the paper uses "GDFusion"; unify the terminology.

Circularity Check

1 steps flagged · score 4.0 of 10

One self-definitional equivalence drives the 'unified gradient descent' framing; the benchmark comparisons stand independently.

  1. self definitional [Sec. 4.1, Proposition 1 (and proof); see also voxel fusion Eq. (23)]
    "Proposition 1. The RNN update step ht = Aht−1 +Bxt is equivalent to a gradient descent step on ht−1 minimizing the loss function Lt = ∥Aht−1−Bxt∥2. ... Let A′ =I− 2ηA⊤A and B′ = 2ηA⊤B, then: ht =A′ht−1 +B′xt. matching the RNN update form, proving the equivalence."

    The loss is defined using the very matrices A and B whose recurrence it is supposed to explain, so the equivalence holds by construction rather than by independent derivation. Any affine recurrence h' = P h + Q x can be repackaged as a gradient step by choosing a quadratic whose gradient is (P−I)h + Q x; the construction carries no information about why this fusion should work. The proof also silently renames the RNN matrices to A' = I − 2ηA^T A and B' = 2ηA^T B, so the gradient-descent step is not literally the original RNN update but a reparameterized one. The subsequent voxel module (Eq. 23) is simply the vanilla RNN stated directly, so the 'gradient descent view' is a post hoc label rather than a derivation.

full rationale

Most of the paper is self-contained against external benchmarks: the fusion modules are explicit update rules (Eqs. 12, 17, 21, 23), trained with the standard occupancy objective plus module-specific self-supervised losses, and evaluated on Occ3D, SurroundOcc, and OpenOccupancy. The uniform 12-epoch protocol for reimplemented baselines is a measurement-fairness concern, not a circularity one, and the omission of FB-Occ on SurroundOcc removes one confirmation without making the remaining claims circular. Self-citations such as ALOcc [6] are used as baseline material and related work, not as the load-bearing justification for the proposed updates. The geometry, motion, and scene losses do lead to their stated updates by direct differentiation, so those derivations are honest even if the 'unified optimization' framing is cosmetic. The one genuine circular element is Proposition 1: the loss is assembled from the same matrices whose RNN update it purports to derive, and the proof renames those matrices, so the claimed equivalence is true by definition. This weakens the theoretical unification narrative but does not invalidate the update rules or the empirical comparisons, which would stand even if the gradient-descent story were removed. Score 4 reflects partial circularity in the central framing, not in the benchmark evidence.

Assumptions & free parameters 6 free parameters · 6 assumptions · 1 invented entities

The method introduces several hand-chosen hyperparameters (eta_s, eta_m) and a set of learnable projections and transition matrices. The key domain assumptions are temporal consistency of scene and geometry, plus the fairness of the uniform baseline training protocol. One latent-state entity (scene-adaptive parameters) is introduced as a new memory container. No new physical entities, forces, or dimensions are postulated.

free parameters (6)
  • eta_s (scene-level learning rate) = 0.1 (selected; range 0.1-100 tested, mIoU 42.1-42.5)
    Hand-chosen update magnitude for scene-adaptive parameter updates; sensitivity analysis in Table A.14 shows low impact.
  • eta_m (motion learning rate) = 0.01 (range 0.001-0.1 tested, mIoU ~42.4-42.5)
    Hand-chosen learning rate for motion refinement; Table A.13 shows robustness across a 100x range.
  • Q1, Q2, Qo (augmentation/output projections) = learned (c x c matrices)
    Learnable linear projections used in the scene-level self-supervised loss and in the final feature injection; trained end-to-end.
  • S^0 (initial scene parameters) = learned (gamma0, beta0, W0, b0)
    Initial layer-norm scale/shift and linear weights, optimized jointly during training in a meta-learning style.
  • A_v, B_v (voxel RNN matrices) = learned (c x c matrices)
    Transition and input matrices for the single-frame voxel-level RNN state in Eq. 23.
  • f_m, f_eta_g (motion estimator and gating nets) = learned parameters
    Linear layer for motion prediction and the small network computing the adaptive geometry gating scalar.
assumptions (6)
  • standard math Proposition 1: a linear RNN update h_t = A h_{t-1} + B x_t is equivalent to a gradient descent step on ||A h_{t-1} - B x_t||^2.
    Used as the theoretical basis for all four fusion rules. The proof in Sec 4.1 shows gradient descent yields an RNN with transformed matrices A' and B', not the same A and B, so the stated equivalence is not fully established as written.
  • domain assumption Scene conditions (lighting, weather, road type) are consistent over short time spans.
    Motivates the scene-level fusion module in Sec 3.2; if false, test-time adaptation of scene parameters would track noise rather than stable scene properties.
  • domain assumption Scene geometry is consistent over time and can be averaged across frames.
    Underlies the geometry fusion EMA in Eqs. 18-21; assumes depth distributions from adjacent frames describe the same static structure.
  • ad hoc to paper Random linear projections Q1 and Q2 preserve enough information to learn scene-level attributes via the self-supervised loss.
    The loss in Eq. 7 uses linear augmentations; no argument is given that linear projections capture lighting, weather, or road type beyond what the network learns on its own.
  • domain assumption Frame-by-frame gradient updates of scene parameters remain stable during inference without catastrophic forgetting or drift.
    The inference-time updates in Eq. 12 assume repeated gradient steps on a stream of frames do not destabilize the network; only empirical ablations support this.
  • domain assumption The reimplemented baselines are faithful representations of the original methods under a common 12-epoch CBGS training protocol.
    The performance comparison in Sec 5 assumes that a uniform training schedule is fair to all baselines, which may not match each baseline's original training recipe.
invented entities (1)
  • Scene-adaptive parameter state S_t (gamma, beta, W, b)
    purpose: Stores global scene conditions (lighting, weather, road type) in network parameters, updated frame-by-frame at inference via gradient descent on a self-supervised loss.
    The paper's ablations show that updating these parameters improves mIoU, but there is no external, independent measurement of what S_t encodes or that it generalizes beyond the nuScenes benchmarks.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Rethinking Temporal Fusion with a Unified Gradient Descent View for 3D Semantic Occupancy Prediction." pith.science (2026). https://pith.science/paper/POGCPCPS

@misc{pith2026250412959,
  author       = {Pith},
  title        = {Pith review of: Rethinking Temporal Fusion with a Unified Gradient Descent View for 3D Semantic Occupancy Prediction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/POGCPCPS}},
  note         = {Machine review of arXiv:2504.12959}
}
read the original abstract

We present GDFusion, a temporal fusion method for vision-based 3D semantic occupancy prediction (VisionOcc). GDFusion opens up the underexplored aspects of temporal fusion within the VisionOcc framework, focusing on both temporal cues and fusion strategies. It systematically examines the entire VisionOcc pipeline, identifying three fundamental yet previously overlooked temporal cues: scene-level consistency, motion calibration, and geometric complementation. These cues capture diverse facets of temporal evolution and make distinct contributions across various modules in the VisionOcc framework. To effectively fuse temporal signals across heterogeneous representations, we propose a novel fusion strategy by reinterpreting the formulation of vanilla RNNs. This reinterpretation leverages gradient descent on features to unify the integration of diverse temporal information, seamlessly embedding the proposed temporal cues into the network. Extensive experiments on nuScenes demonstrate that GDFusion significantly outperforms established baselines. Notably, on Occ3D benchmark, it achieves 1.4\%-4.8\% mIoU improvements and reduces memory consumption by 27\%-72\%.

Figures

Figures reproduced from arXiv: 2504.12959 by the authors.

Figure 1
Figure 1. Motivation behind the proposed temporal fusion. (a): VisionOcc pipeline. (b): Proposed temporal cues, showing historical motion and geometric data enhancing current viewpoints, with scene consistency priors from historical information. etry cue: Advanced VisionOcc methods utilize geometric priors (e.g., depth) [6, 23, 31, 39] for 2D-to-3D lifting. His￾torical geometric information could effectively compensate for th… view at source ↗
Figure 2
Figure 2. Multi-level temporal fusion in the VisionOcc pipeline. Volume features Vt , geometry Gt , motion Mt , and scene-adaptive parameters S t are enhanced through RNN-style temporal fusion, each capturing distinct temporal dynamics. Single-frame-sized historical states Ht−1 v , Ht−1 g , Ht−1 m , and Ht−1 s are stored in memory and updated frame-by-frame. [15, 49], this information is expected to capture abstract, domain-s… view at source ↗
Figure 3
Figure 3. Update dynamics of gradient descent-based temporal fusion pipeline. f t denotes the (geometry, motion, voxel-level, scene-level) feature of the current frame. Ht−1 and Ht represent the prior and current historical states, respectively. Let A′ = I − 2ηA⊤A and B′ = 2ηA⊤B, then: h t = A ′h t−1 + B ′x t . matching the RNN update form, proving the equivalence. Next, we will detail how Prop. 1 guides the fusion of distinc… view at source ↗
Figures from the paper (1 more)
Figure 5
Figure 5. Figure 5: Comparison of our GDFusion and SOLOFusion w.r.t. memory consumption. SOLOFusion boosts performance with longer sequences but increases inference memory, while GDFu￾sion achieves high performance with low memory consumption. backbone with 256 × 704 image inputs for most…

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. SHTOcc: Effective 3D Occupancy Prediction with Sparse Head and Tail Voxels

    cs.CV 2025-05 reject novelty 5.0 of 10

    SHTOcc combines attention-based sparse voxel selection with decoupled classifier retraining for 3D occupancy prediction, reporting efficiency gains and small, partly inconsistent accuracy improvements.

Reference graph

Works this paper leans on

72 extracted references · 40 canonical work pages · cited by 1 Pith paper

  1. [1]

    How to train your maml

    Antreas Antoniou, Harrison Edwards, and Amos Storkey. How to train your maml. In Proceedings of International Conference on Learning Representations, 2018. 5

    work page 2018

  2. [2]

    xl- stm: Extended long short-term memory

    Maximilian Beck, Korbinian P ¨oppel, Markus Spanring, An- dreas Auer, Oleksandra Prudnikova, Michael Kopp, G ¨unter Klambauer, Johannes Brandstetter, and Sepp Hochreiter. xl- stm: Extended long short-term memory. arXiv preprint arXiv:2405.04517, 2024. 3

    arXiv 2024

  3. [3]

    Ti- tans: Learning to memorize at test time

    Ali Behrouz, Peilin Zhong, and Vahab Mirrokni. Ti- tans: Learning to memorize at test time. arXiv preprint arXiv:2501.00663, 2024. 3, 5

    arXiv 2024

  4. [4]

    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, 2020. 7

    work page 2020

  5. [5]

    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, 2022. 3

    work page 2022

  6. [6]

    Alocc: adaptive lifting-based 3d semantic occu- pancy and cost volume-based flow prediction.arXiv preprint arXiv:2411.07725, 2024

    Dubing Chen, Jin Fang, Wencheng Han, Xinjing Cheng, Junbo Yin, Chengzhong Xu, Fahad Shahbaz Khan, and Jian- bing Shen. Alocc: adaptive lifting-based 3d semantic occu- pancy and cost volume-based flow prediction.arXiv preprint arXiv:2411.07725, 2024. 1, 2, 3, 4, 6, 7, 8, 9

    arXiv 2024

  7. [7]

    Evolu- tionary generalized zero-shot learning

    Dubing Chen, Chenyi Jiang, and Haofeng Zhang. Evolu- tionary generalized zero-shot learning. In Proceedings of the Thirty-Third International Joint Conference on Artificial In- telligence, IJCAI-24, pages 632–640, 2024. 2, 4, 5

    work page 2024

  8. [8]

    Exploring simple siamese rep- resentation learning

    Xinlei Chen and Kaiming He. Exploring simple siamese rep- resentation learning. In Proceedings of the IEEE/CVF con- ference on computer vision and pattern recognition , pages 15750–15758, 2021. 5

    2021

Show all 72 references
  1. [9]

    3d sketch-aware semantic scene comple- tion via semi-supervised structure prior

    Xiaokang Chen, Kwan-Yee Lin, Chen Qian, Gang Zeng, and Hongsheng Li. 3d sketch-aware semantic scene comple- tion via semi-supervised structure prior. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 4193–4202, 2020. 9, 4

    2020

  2. [10]

    Learning phrase representations using rnn encoder-decoder for statistical machine translation

    Kyunghyun Cho. Learning phrase representations using rnn encoder-decoder for statistical machine translation. In Pro- ceedings of Empirical Methods in Natural Language Pro- cessing, 2014. 6

    2014

  3. [11]

    Deformable convolutional networks

    Jifeng Dai, Haozhi Qi, Yuwen Xiong, Yi Li, Guodong Zhang, Han Hu, and Yichen Wei. Deformable convolutional networks. In Proceedings of the IEEE International Confer- ence on Computer Vision, pages 764–773, 2017. 4, 6

    2017

  4. [12]

    Using occupancy grids for mobile robot per- ception and navigation

    Alberto Elfes. Using occupancy grids for mobile robot per- ception and navigation. Computer, 1989. 1

    1989

  5. [13]

    Model- agnostic meta-learning for fast adaptation of deep networks

    Chelsea Finn, Pieter Abbeel, and Sergey Levine. Model- agnostic meta-learning for fast adaptation of deep networks. In Proceedings of International Conference on Machine Learning, 2017. 5

    2017

  6. [14]

    Gaussianocc: Fully self-supervised and effi- cient 3d occupancy estimation with gaussian splatting.arXiv preprint arXiv:2408.11447, 2024

    Wanshui Gan, Fang Liu, Hongbin Xu, Ningkai Mo, and Naoto Yokoya. Gaussianocc: Fully self-supervised and effi- cient 3d occupancy estimation with gaussian splatting.arXiv preprint arXiv:2408.11447, 2024. 3

    2024 arXiv

  7. [15]

    Unsupervised domain adaptation by backpropagation

    Yaroslav Ganin and Victor Lempitsky. Unsupervised domain adaptation by backpropagation. In International conference on machine learning, pages 1180–1189. PMLR, 2015. 4

    2015

  8. [16]

    Learning fast approxima- tions of sparse coding

    Karol Gregor and Yann LeCun. Learning fast approxima- tions of sparse coding. In Proceedings of the 27th inter- national conference on international conference on machine learning, pages 399–406, 2010. 3

    2010

  9. [17]

    Mamba: Linear-time sequence modeling with selective state spaces

    Albert Gu and Tri Dao. Mamba: Linear-time sequence modeling with selective state spaces. arXiv preprint arXiv:2312.00752, 2023. 3

    2023 arXiv

  10. [18]

    Exploring recurrent long-term temporal fusion 10 for multi-view 3d perception

    Chunrui Han, Jinrong Yang, Jianjian Sun, Zheng Ge, Run- pei Dong, Hongyu Zhou, Weixin Mao, Yuang Peng, and Xi- angyu Zhang. Exploring recurrent long-term temporal fusion 10 for multi-view 3d perception. IEEE Robotics and Automa- tion Letters, 2024. 1, 3, 6, 7

    2024

  11. [19]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2016. 7

    2016

  12. [20]

    Fas- tocc: Accelerating 3d occupancy prediction by fusing the 2d bird’s-eye view and perspective view

    Jiawei Hou, Xiaoyan Li, Wenhao Guan, Gang Zhang, Di Feng, Yuheng Du, Xiangyang Xue, and Jian Pu. Fas- tocc: Accelerating 3d occupancy prediction by fusing the 2d bird’s-eye view and perspective view. In IEEE International Conference on Robotics and Automation, 2024. 8

    2024

  13. [21]

    St-p3: End-to-end vision-based au- tonomous driving via spatial-temporal feature learning

    Shengchao Hu, Li Chen, Penghao Wu, Hongyang Li, Junchi Yan, and Dacheng Tao. St-p3: End-to-end vision-based au- tonomous driving via spatial-temporal feature learning. In Proceedings of European Conference on Computer Vision ,

  14. [22]

    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, 2023. 1

    2023

  15. [23]

    Bevdet4d: Exploit tempo- ral cues in multi-camera 3d object detection

    Junjie Huang and Guan Huang. Bevdet4d: Exploit tempo- ral cues in multi-camera 3d object detection. arXiv preprint arXiv:2203.17054, 2022. 1, 2, 3, 6, 7, 8, 4

    2022 arXiv

  16. [24]

    Bevdet: High-performance multi-camera 3d object de- tection in bird-eye-view

    Junjie Huang, Guan Huang, Zheng Zhu, Yun Ye, and Dalong Du. Bevdet: High-performance multi-camera 3d object de- tection in bird-eye-view. arXiv preprint arXiv:2112.11790,

    arXiv

  17. [25]

    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, 2023. 1, 3, 8, 9, 4

    2023

  18. [26]

    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. InEuropean Conference on Computer Vision , pages 376–393. Springer,

  19. [27]

    Vad: Vectorized scene representation 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 representation for efficient autonomous driving. In Proceedings of the In- ternational Conference on Computer Vision, 2023. 1

    2023

  20. [28]

    Anisotropic convolutional networks for 3d semantic scene completion

    Jie Li, Kai Han, Peng Wang, Yu Liu, and Xia Yuan. Anisotropic convolutional networks for 3d semantic scene completion. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 3351– 3359, 2020. 9, 4

    2020

  21. [29]

    Viewformer: Exploring spatiotem- poral modeling for multi-view 3d occupancy perception via view-guided transformers

    Jinke Li, Xiao He, Chonghua Zhou, Xiaoqiang Cheng, Yang Wen, and Dan Zhang. Viewformer: Exploring spatiotem- poral modeling for multi-view 3d occupancy perception via view-guided transformers. In Proceedings of European Con- ference on Computer Vision, 2024. 3, 7, 8

    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, Qiao Yu, and Jifeng Dai. Bevformer: Learning bird’s-eye-view representation from multi-camera images via spatiotemporal transformers. arXiv preprint arXiv:2203.17270, 2022. 3, 8

    2022 arXiv

  23. [31]

    Fb-occ: 3d occupancy prediction based on forward-backward view transformation

    Zhiqi Li, Zhiding Yu, David Austin, Mingsheng Fang, Shiyi Lan, Jan Kautz, and Jose M Alvarez. Fb-occ: 3d occupancy prediction based on forward-backward view transformation. arXiv preprint arXiv:2307.01492, 2023. 1, 2, 3, 4, 6, 7, 8, 9

    2023 arXiv

  24. [32]

    Sparse4d v2: Recurrent temporal fusion with sparse model

    Xuewu Lin, Tianwei Lin, Zixiang Pei, Lichao Huang, and Zhizhong Su. Sparse4d v2: Recurrent temporal fusion with sparse model. arXiv preprint arXiv:2305.14018, 2023. 3

    2023 arXiv

  25. [33]

    Fully sparse 3d occupancy prediction

    Haisong Liu, Haiguang Wang, Yang Chen, Zetong Yang, Jia Zeng, Li Chen, and Limin Wang. Fully sparse 3d occupancy prediction. In Proceedings of European Conference on Com- puter Vision, 2024. 3, 4

    2024

  26. [34]

    Surroundsdf: Implicit 3d scene understanding based on signed distance field

    Lizhe Liu, Bohua Wang, Hongwei Xie, Daqi Liu, Li Liu, Zhiqiang Tian, Kuiyuan Yang, and Bing Wang. Surroundsdf: Implicit 3d scene understanding based on signed distance field. In Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition, 2024. 1, 8, 3

    2024

  27. [35]

    Petr: Position embedding transformation for multi-view 3d object detection

    Yingfei Liu, Tiancai Wang, Xiangyu Zhang, and Jian Sun. Petr: Position embedding transformation for multi-view 3d object detection. arXiv preprint arXiv:2203.05625, 2022. 3

    2022 arXiv

  28. [36]

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

    Yingfei Liu, Junjie Yan, Fan Jia, Shuailin Li, Qi Gao, Tian- cai Wang, Xiangyu Zhang, and Jian Sun. Petrv2: A uni- fied framework for 3d perception from multi-camera images. arXiv preprint arXiv:2206.01256, 2022. 3

    2022 arXiv

  29. [37]

    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 (ICCV), 2021. 7

    2021

  30. [38]

    Oc- treeocc: Efficient and multi-granularity occupancy predic- tion using octree queries

    Yuhang Lu, Xinge Zhu, Tai Wang, and Yuexin Ma. Oc- treeocc: Efficient and multi-granularity occupancy predic- tion using octree queries. arXiv preprint arXiv:2312.03774,

    arXiv

  31. [39]

    Cotr: Compact occupancy transformer for vision-based 3d occupancy prediction

    Qihang Ma, Xin Tan, Yanyun Qu, Lizhuang Ma, Zhizhong Zhang, and Yuan Xie. Cotr: Compact occupancy transformer for vision-based 3d occupancy prediction. arXiv preprint arXiv:2312.01919, 2023. 1, 2, 3, 6, 7, 8

    2023 arXiv

  32. [40]

    Co-occ: Coupling explicit feature fusion with volume rendering regularization for multi-modal 3d semantic occupancy prediction

    Jingyi Pan, Zipeng Wang, and Lin Wang. Co-occ: Coupling explicit feature fusion with volume rendering regularization for multi-modal 3d semantic occupancy prediction. IEEE Robotics and Automation Letters, 2024. 9, 4

    2024

  33. [41]

    Uniocc: Unifying vision-centric 3d oc- cupancy prediction with geometric and semantic rendering

    Mingjie Pan, Li Liu, Jiaming Liu, Peixiang Huang, Lon- glong Wang, Shanghang Zhang, Shaoqing Xu, Zhiyi Lai, and Kuiyuan Yang. Uniocc: Unifying vision-centric 3d oc- cupancy prediction with geometric and semantic rendering. arXiv preprint arXiv:2306.09117, 2023. 3, 8

    2023 arXiv

  34. [42]

    Renderocc: Vision-centric 3d occupancy prediction with 2d rendering supervision

    Mingjie Pan, Jiaming Liu, Renrui Zhang, Peixiang Huang, Xiaoqi Li, Hongwei Xie, Bing Wang, Li Liu, and Shanghang Zhang. Renderocc: Vision-centric 3d occupancy prediction with 2d rendering supervision. In Proceedings of IEEE In- ternational Conference on Robotics and Automation...

    2024

  35. [43]

    Time will tell: New outlooks and a baseline for temporal multi-view 3d object detection

    Jinhyung Park, Chenfeng Xu, Shijia Yang, Kurt Keutzer, Kris Kitani, Masayoshi Tomizuka, and Wei Zhan. Time will tell: New outlooks and a baseline for temporal multi-view 3d object detection. arXiv preprint arXiv:2210.02443, 2022. 1, 2, 3, 6, 7, 8, 9, 4

    arXiv

  36. [44]

    Rwkv: Reinventing rnns for the transformer era

    Bo Peng, Eric Alcaide, Quentin Anthony, Alon Albalak, Samuel Arcadinho, Stella Biderman, Huanqi Cao, Xin 11 Cheng, Michael Chung, Matteo Grella, et al. Rwkv: Reinventing rnns for the transformer era. arXiv preprint arXiv:2305.13048, 2023. 3

    2023 arXiv

  37. [45]

    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

  38. [46]

    Recurrent inference machines for solving inverse problems

    Patrick Putzky and Max Welling. Recurrent inference machines for solving inverse problems. arXiv preprint arXiv:1706.04008, 2017. 3

    2017 arXiv

  39. [47]

    The devil in linear transformer

    Zhen Qin, Xiaodong Han, Weixuan Sun, Dongxu Li, Ling- peng Kong, Nick Barnes, and Yiran Zhong. The devil in linear transformer. arXiv preprint arXiv:2210.10340, 2022. 5

    2022 arXiv

  40. [48]

    Occupancy as set of points

    Yiang Shi, Tianheng Cheng, Qian Zhang, Wenyu Liu, and Xinggang Wang. Occupancy as set of points. In Computer Vision–ECCV 2024: 18th European Conference, 2024. 3, 8

    2024

  41. [49]

    Test-time training with self- supervision for generalization under distribution shifts

    Yu Sun, Xiaolong Wang, Zhuang Liu, John Miller, Alexei Efros, and Moritz Hardt. Test-time training with self- supervision for generalization under distribution shifts. In Proceedings of the 37th International Conference on Ma- chine Learning, pages 9229–9248, 2020. 2, 4, 5

    2020

  42. [50]

    Learning to (learn at test time): Rnns with expressive hidden states

    Yu Sun, Xinhao Li, Karan Dalal, Jiarui Xu, Arjun Vikram, Genghan Zhang, Yann Dubois, Xinlei Chen, Xiaolong Wang, Sanmi Koyejo, et al. Learning to (learn at test time): Rnns with expressive hidden states. arXiv preprint arXiv:2407.04620, 2024. 2, 3, 4, 5

    2024 arXiv

  43. [51]

    Occ3d: A large-scale 3d occupancy prediction benchmark for autonomous driving

    Xiaoyu Tian, Tao Jiang, Longfei Yun, Yucheng Mao, Huitong Yang, Yue Wang, Yilun Wang, and Hang Zhao. Occ3d: A large-scale 3d occupancy prediction benchmark for autonomous driving. In Proceedings of Advances in Neu- ral Information Processing Systems, 2024. 1, 3, 7

    2024

  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, 2023. 1

    2023

  45. [53]

    Attention is all you need

    A Vaswani. Attention is all you need. Advances in Neural Information Processing Systems, 2017. 9

    2017

  46. [54]

    Occgen: Gener- ative multi-modal 3d occupancy prediction for autonomous driving

    Guoqing Wang, Zhongdao Wang, Pin Tang, Jilai Zheng, Xi- angxuan Ren, Bailan Feng, and Chao Ma. Occgen: Gener- ative multi-modal 3d occupancy prediction for autonomous driving. In European Conference on Computer Vision, pages 95–112. Springer, 2024. 9, 4

    2024

  47. [55]

    Opus: occupancy prediction using a sparse set.arXiv preprint arXiv:2409.09350, 2024

    Jiabao Wang, Zhaojiang Liu, Qiang Meng, Liujiang Yan, Ke Wang, Jie Yang, Wei Liu, Qibin Hou, and Mingming Cheng. Opus: occupancy prediction using a sparse set.arXiv preprint arXiv:2409.09350, 2024. 3, 8, 4

    2024 arXiv

  48. [56]

    Occsora: 4d occupancy generation models as world simulators for au- tonomous driving

    Lening Wang, Wenzhao Zheng, Yilong Ren, Han Jiang, Zhiyong Cui, Haiyang Yu, and Jiwen Lu. Occsora: 4d occupancy generation models as world simulators for au- tonomous driving. arXiv preprint arXiv:2405.20337, 2024. 3

    2024 arXiv

  49. [57]

    Exploring object-centric temporal modeling for efficient multi-view 3d object detection

    Shihao Wang, Yingfei Liu, Tiancai Wang, Ying Li, and Xi- angyu Zhang. Exploring object-centric temporal modeling for efficient multi-view 3d object detection. In In Proceed- ings of the IEEE/CVF international conference on computer vision, pages 3621–3631, 2023. 1, 3, 6, 7

    2023

  50. [58]

    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. arXiv preprint arXiv:2303.03991, 2023. 1, 3, 7, 9, 4

    2023 arXiv

  51. [59]

    Panoocc: Unified occupancy representation for camera-based 3d panoptic segmentation

    Yuqi Wang, Yuntao Chen, Xingyu Liao, Lue Fan, and Zhaox- iang Zhang. Panoocc: Unified occupancy representation for camera-based 3d panoptic segmentation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024. 8

    2024

  52. [60]

    Occllama: An occupancy-language-action generative world model for au- tonomous driving, 2024

    Julong Wei, Shanshuai Yuan, Pengfei Li, Qingda Hu, Zhongxue Gan, and Wenchao Ding. Occllama: An occupancy-language-action generative world model for au- tonomous driving, 2024. 3

    2024

  53. [61]

    Surroundocc: Multi-camera 3d oc- cupancy prediction for autonomous driving

    Yi Wei, Linqing Zhao, Wenzhao Zheng, Zheng Zhu, Jie Zhou, and Jiwen Lu. Surroundocc: Multi-camera 3d oc- cupancy prediction for autonomous driving. arXiv preprint arXiv:2303.09551, 2023. 3, 7, 8

    2023 arXiv

  54. [62]

    Bevformer v2: Adapting modern image backbones to bird’s-eye-view recognition via perspective su- pervision

    Chenyu Yang, Yuntao Chen, Hao Tian, Chenxin Tao, Xizhou Zhu, Zhaoxiang Zhang, Gao Huang, Hongyang Li, Yu Qiao, Lewei Lu, et al. Bevformer v2: Adapting modern image backbones to bird’s-eye-view recognition via perspective su- pervision. In Proceedings of the IEEE/CVF Conference...

    2023

  55. [63]

    Flashocc: Fast and memory-efficient occupancy prediction via channel-to-height plugin

    Zichen Yu, Changyong Shu, Jiajun Deng, Kangjie Lu, Zong- dai Liu, Jiangyong Yu, Dawei Yang, Hui Li, and Yan Chen. Flashocc: Fast and memory-efficient occupancy prediction via channel-to-height plugin. arXiv preprint arXiv:2311.12058, 2023. 3, 7, 8, 9

    2023 arXiv

  56. [64]

    Panoptic-flashocc: An efficient baseline to marry semantic occupancy with panoptic via instance cen- ter

    Zichen Yu, Changyong Shu, Qianpu Sun, Junjie Linghu, Xi- aobao Wei, Jiangyong Yu, Zongdai Liu, Dawei Yang, Hui Li, and Yan Chen. Panoptic-flashocc: An efficient baseline to marry semantic occupancy with panoptic via instance cen- ter. arXiv preprint arXiv:2406.10527, 2024. 4

    2024 arXiv

  57. [65]

    Streammapnet: Streaming mapping network for vectorized online hd map construction

    Tianyuan Yuan, Yicheng Liu, Yue Wang, Yilun Wang, and Hang Zhao. Streammapnet: Streaming mapping network for vectorized online hd map construction. In In Proceedings of the IEEE/CVF Winter Conference on Applications of Com- puter Vision, pages 7356–7365, 2024. 7

    2024

  58. [66]

    Occnerf: Self- supervised multi-camera occupancy prediction with neural radiance fields

    Chubin Zhang, Juncheng Yan, Yi Wei, Jiaxin Li, Li Liu, Yansong Tang, Yueqi Duan, and Jiwen Lu. Occnerf: Self- supervised multi-camera occupancy prediction with neural radiance fields. arXiv preprint arXiv:2312.09243, 2023. 3

    2023 arXiv

  59. [67]

    Occloff: Learning op- timized feature fusion for 3d occupancy prediction

    Ji Zhang, Yiran Ding, and Zixin Liu. Occloff: Learning op- timized feature fusion for 3d occupancy prediction. In 2025 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), pages 3096–3106. IEEE, 2025. 9, 4

    2025

  60. [68]

    Vision-based 3d occupancy prediction in autonomous driving: a review and outlook

    Yanan Zhang, Jinqing Zhang, Zengran Wang, Junhao Xu, and Di Huang. Vision-based 3d occupancy prediction in autonomous driving: a review and outlook. arXiv preprint arXiv:2405.02595, 2024. 3

    2024 arXiv

  61. [69]

    Conditional random fields as 12 recurrent neural networks

    Shuai Zheng, Sadeep Jayasumana, Bernardino Romera- Paredes, Vibhav Vineet, Zhizhong Su, Dalong Du, Chang Huang, and Philip HS Torr. Conditional random fields as 12 recurrent neural networks. In Proceedings of the IEEE inter- national conference on computer vision , pages 1529–1537,

  62. [70]

    Occworld: Learning a 3d occupancy world model for autonomous driving

    Wenzhao Zheng, Weiliang Chen, Yuanhui Huang, Borui Zhang, Yueqi Duan, and Jiwen Lu. Occworld: Learning a 3d occupancy world model for autonomous driving. In European Conference on Computer Vision , pages 55–72. Springer, 2025. 3

    2025

  63. [71]

    Deformable detr: Deformable transformers for end-to-end object detection

    Xizhou Zhu, Weijie Su, Lewei Lu, Bin Li, Xiaogang Wang, and Jifeng Dai. Deformable detr: Deformable transformers for end-to-end object detection. In Proceedings of Interna- tional Conference on Learning Representations, 2020. 4, 6

    2020

  64. [72]

    Gaussianworld: Gaussian world model for streaming 3d occupancy prediction

    Sicheng Zuo, Wenzhao Zheng, Yuanhui Huang, Jie Zhou, and Jiwen Lu. Gaussianworld: Gaussian world model for streaming 3d occupancy prediction. arXiv preprint arXiv:2412.10373, 2024. 8, 3 13 A. Derivation of the Gradients A.1. Gradient Computation for Scene-Level Temporal Fusion...

    2024 arXiv

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.