CoverPrune: Coverage-Driven Token Pruning for 3D VLMs via Optimal Transport

Lingting Zhu; Peng Ling; Shengju Qian; Weikai Chen; Wenming Yang; Xin Wang; Yingda Yin; Zeyu Hu

REVIEW 2 major objections 6 minor 42 references

CoverPrune: Coverage-Driven Token Pruning for 3D VLMs via Optimal Transport

T0 review · 2 major / 6 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read Pruning visual tokens by coverage, formalized as optimal transport, preserves spatial reasoning under extreme compression: CoverPrune keeps 92.4% of full-token VSI-Bench performance at 20% retention and beats attention- and…

desk verdict Strong empirical token pruning results for 3D VLMs, but the SGS algorithm is degenerate as stated and the theoretical guarantees collapse; the Lite variant is the cleaner contribution. read the letter →

arxiv 2608.13226 v1 pith:P3KE55YS submitted 2026-08-13 cs.CV cs.AI

Peng Ling , Yingda Yin , Lingting Zhu , Weikai Chen , Shengju Qian , Zeyu Hu , Xin Wang , Wenming Yang This is my paper · ORCID

classification cs.CVcs.AI

keywords visualtokenpruningoptimaltransport3Dvision-languagemodelsspatialreasoningevidencecoverageefficiencyinferenceacceleration

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

Three-dimensional vision-language models gain spatial reasoning by injecting geometry into the visual token stream, but the resulting token counts make inference expensive. CoverPrune argues that the right pruning objective is coverage — keeping a compact set of prototype tokens that collectively explain the full evidence — rather than diversity, which tends to discard representative tokens in favor of outliers. The paper formalizes coverage as an optimal transport problem: the retained tokens serve as prototypes that must carry the mass of the full token set under a cost combining feature similarity, 3D distance, and temporal order, with informative tokens receiving extra target capacity. A semi-relaxed transport formulation and a spatial-guided greedy selection make the intractable subset choice practical, and a block-structured variant runs in near-linear time. If the argument holds, aggressive token pruning can accelerate 3D VLMs without sacrificing spatial reasoning; the reported numbers show 92.4% of full-token VSI-Bench performance retained at 20% of the tokens.

What carries the argument

The central machinery is the coverage-as-OT objective: a semi-relaxed optimal transport problem in which the retained tokens are the source, the original tokens are the target, and the objective is to minimize transport cost while allowing unused target capacity. The cost matrix is the Feature-Spatial-Temporal (FST) cost $C_{ij} = \lambda_f \hat{d}_f + \lambda_x \phi_\kappa(\hat{d}_x) + \lambda_\tau \hat{d}_\tau$, blending normalized feature cosine distance, a log-compressed 3D spatial distance, and a one-sided temporal penalty $\mathrm{ReLU}(\tau_s - \tau_t)$ so a later token cannot cheaply cover an earlier one. Per-token target capacities $v_j$ are reweighted by local FST distinctiveness, concentrating transport mass on tokens that are hard to approximate from their 3D neighbors. The semi-relaxed inequality $P^\top \mathbf{1} \le v$ is the load-bearing relaxation: it turns the combinatorial subset selection into a monotone submodular objective with greedy constant-factor guarantees, and Spatial-Guided Greedy Selection implements the greedy step by evaluating marginal costs only within each candidate's $g$ nearest 3D neighbors. CoverPrune-Lite replaces this iteration with a Morton-code ordering, capacity-balanced grouping into $K$ groups, and one within-group prototype per group, reducing the pruning pass to $O(N \log N)$.

What would settle it

Take the kept sets produced by CoverPrune and by each baseline at the same budget on VSI-Bench, compute the FST transport cost of Eqs. (8)–(10) for each kept set, and compare those costs with per-question accuracy. If the kept set with the lowest transport cost is not the one with the highest accuracy across the benchmark, then the defined coverage objective is not the mechanism producing the pruning gains.

Watch

Extended reading notes

Core claim

The paper's central claim is that coverage, not diversity, is the correct objective for pruning visual tokens in 3D VLMs, and that this objective can be cast as an optimal transport problem with a Feature-Spatial-Temporal cost. In this view, the selected subset $S$ is the source support and the full token set $T$ is the target support; pruning minimizes the cost of transporting the target distribution to $S$, subject to uniform source capacities and an informativeness-weighted target capacity $v$ computed from local FST distinctiveness. The semi-relaxed constraint $P^\top \mathbf{1} \le v$ leaves residual target capacity unused, which is what makes subset selection a monotone submodular maximization admitting greedy approximation. The paper reports that this coverage-based selection outperforms attention-based and diversity-based pruning on Scan2Cap, ScanQA, SQA3D, and VSI-Bench across two base models, and that at 20% token retention it preserves 92.4% of full-token VSI-Bench performance, with the advantage growing as the retention ratio drops to 10% and 5%.

Load-bearing premise

The load-bearing premise is that minimizing the transport cost between the kept tokens and all original tokens, under the paper's feature-spatial-temporal weighting and capacity vector, faithfully tracks whether the retained visual evidence is enough for correct spatial reasoning; the paper checks this only through final benchmark accuracy.

Editorial extensions

If this is right

At 20% retention, CoverPrune keeps 92.4% of full-token VSI-Bench accuracy with the stronger base model, while at 10% and 5% it holds a clear lead over all four baselines, indicating the coverage objective degrades more gracefully under compression.
On general 3D reasoning benchmarks, 10%-retention CoverPrune roughly matches or exceeds the accuracy that diversity-based baselines reach at 20% retention on multiple metrics, so aggressive pruning no longer forces a proportional accuracy loss.
Ablations show that removing the feature term from the FST cost costs about 3.6 overall VSI-Bench points, while removing the geometry and temporal terms costs roughly 0.7–0.8 points each, so semantic affinity is the dominant term but spatial and temporal cues are not noise.
CoverPrune-Lite cuts pruning time from 2.53 seconds to 0.41 seconds on VSI-Bench while slightly improving relative accuracy (88.01% vs 87.84%), making coverage-based pruning usable at deployment latencies.
Because CoverPrune is training-free and plugs in right before the language-model prefill stage, its benefits are available without fine-tuning, which is what the two-base-model experiments demonstrate.

Reading between the lines

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

A direct test the paper does not run: compute the FST transport cost of the kept sets produced by each pruning method and correlate it with accuracy across VSI-Bench questions. If the lowest-cost kept set is not the most accurate, the coverage proxy itself is not the operative mechanism.
The paper's transfer claim is untested beyond two 3D VLMs on indoor-scan benchmarks; a natural extension is outdoor scenes or video-based spatial grounding, where the temporal penalty direction and 3D coordinate estimation would have to be re-examined.
The one-sided temporal cost encodes an assumption about evidence order, namely that earlier tokens should cover later ones. Reversing that direction on the same benchmarks would isolate how much of the gain comes from temporal asymmetry rather than from spatial and feature coverage.
Since the method operates on raw tokens before the prefill stage, it should compose with later-stage accelerations such as KV-cache eviction or attention sparsification; the paper does not measure such compounded speedups.

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.

Desk Editor's Note

read the letter

Two things to know about arXiv:2608.13226. First, the empirical package is solid: across ScanQA, SQA3D, Scan2Cap, and VSI-Bench, with two base models (GS-Reasoner and VLM-3R), CoverPrune and CoverPrune-Lite consistently beat attention- and diversity-based baselines at 10–20% token retention, and the ablations show each FST term contributes. Second, the theoretical core does not survive contact with its own equations. As written, SGS's residual in Eq. (14) is identically zero: u and v are both normalized to sum to 1, so P_l 1 = u forces P_l^T 1 ≤ v with equal total mass, hence equality, and every candidate scores zero in Eq. (15). The paper never states the alternative (partial mass, 1/K per selected token), and if that were intended, the greedy rule is no longer standard submodular maximization, so the cited approximation guarantees from [17,21] do not apply. Submodularity of the FST-induced set function is also asserted, not proved, for a cost with an asymmetric ReLU temporal term.

What is genuinely new: the coverage framing for 3D VLM pruning, the FST transport cost, the informativeness-aware capacity, and the Lite variant's Morton-order grouping with capacity-balanced blocks. These are real contrasts to diversity-based methods. The OT prototype machinery is prior work (SPOT, Partial Wasserstein Covering), and the paper cites it honestly. The reported gains are a few points on average, not a breakthrough, but consistent.

Minor soft spots: hyperparameters like κ, n, and g in the FST cost and capacity are never given, and no code is attached, so the numbers are not independently reproducible. The ablations report only point estimates without variance across runs. Some relative accuracy numbers exceed 100%, suggesting pruning occasionally regularizes; the paper does not comment on that.

The central argument — minimizing OT coverage loss preserves the evidence the VLM needs — is plausible but demonstrated only indirectly through downstream accuracy. That is acceptable for a systems paper, but it means the theory should be either fixed or softened. The Lite variant is conceptually cleaner than full SGS and likely carries the practical value.

Reading group? Yes — it is a nice example of a paper whose empirical results can outrun its own derivations. I would not cite the theoretical claims, and I would wait for a revised draft with the SGS normalization fixed and the missing hyperparameters. Recommendation: send it to peer review. The empirical direction is useful enough that a serious referee should engage, but the authors need to fix SGS, prove or drop the submodularity claim, and release the code and hyperparameters before acceptance.

Referee Report

2 major / 6 minor

Summary. The paper proposes CoverPrune, a training-free inference-time token pruning method for 3D vision-language models (3D VLMs). The method formulates pruning as an optimal transport (OT) problem that selects a compact subset of visual tokens maximizing coverage of the full token set under a feature-spatial-temporal (FST) cost and an informativeness-aware target capacity. To make the combinatorial selection tractable, the paper introduces a Spatial-Guided Greedy Selection (SGS) algorithm that greedily builds the subset using a residual-weighted local cost, and a faster variant, CoverPrune-Lite, based on Morton-order grouping. Experiments on ScanQA, SQA3D, Scan2Cap, and VSI-Bench with two base models (GS-Reasoner and VLM-3R) report consistent improvements over attention- and diversity-based baselines, especially under aggressive pruning (10% and 5% retention).

Significance. If the approach holds, it offers a new coverage-based paradigm for token pruning in 3D VLMs, moving beyond diversity- and attention-based heuristics, and it is training-free and plug-and-play. The empirical evaluation is a strength: results are reported across two base models, four benchmarks, and multiple retention ratios, with external benchmarks and no apparent fitting to test labels. The ablation study isolates the contributions of the FST cost terms and the capacity weighting. The Lite variant provides a practical low-overhead approximation. However, the theoretical justification of the SGS algorithm is a major weakness: the semi-relaxed OT formulation as written degenerates under the stated capacity normalization, and the submodularity/approximation claims are not established for the asymmetric, sample-normalized FST cost.

major comments (2)

[§3.5, Eq. (14)] Under the normalization stated in §3.2 and confirmed in §4.2, both capacity vectors u and v sum to 1. The semi-relaxed problem in Eq. (11) imposes P1 = u and P^T 1 ≤ v. Since the total transported mass is 1 and v sums to 1, the inequality must be tight: P^T 1 = v exactly. Consequently the residual r_ℓ = [v − P_ℓ^T 1]_+ in Eq. (14) is identically zero at every greedy step, and the selection rule in Eq. (15) assigns a score of zero to every candidate token, making SGS arbitrary. If the authors intended u to carry partial mass (for example u_i = 1/K for each selected token, so that the total mass is ℓ/K at step ℓ), this must be stated explicitly, because it contradicts the normalization in §3.2 and §4.2, and the cited approximation guarantees from [17,21] would need to be rederived for that setting. As written, the central algorithm cannot be expected to produce the reported results without additional unspecified implementation details.
[§3.5, third paragraph] The claim that the induced set function for subset selection under the relaxed Wasserstein objective is monotone submodular and therefore admits greedy constant-factor approximation is not established for the FST cost in Eq. (8). The temporal term ReLU(τ_s − τ_t) is asymmetric, and each discrepancy is min-max normalized within the sample, so the cost matrix is not a symmetric metric; the cited results [17,21] do not directly apply. Moreover, the SGS rule in Eq. (15) is not the standard greedy marginal-gain update for submodular maximization: it replaces the global marginal gain with a local residual-weighted sum over a 3D neighborhood. Even if the full OT objective were submodular, the approximation guarantee would not transfer to this local heuristic. The authors should provide a rigorous proof of submodularity for their specific cost or explicitly characterize SGS as a heuristic without theoretical guarantees.

minor comments (6)

[Table 1 caption] The header 'Acc.%↑B-4↑Rouge↑CIDEr↑Meteor↑' contains a typo: 'B-4' should be 'BLEU-4', and the repeated 'Acc.%' labels are confusing. Please clean up the table formatting.
[Table 5] The rows for DTC, CoverPrune, and CoverPrune-Lite are run together with missing spaces (e.g., '62840.73.4723.879.85'), making the efficiency numbers difficult to read. Please fix the LaTeX/table formatting.
[§5.2] The sentence 'At 20% token retention, CoverPrune preserves 92.4% of full-token performance' is supported by Table 2 (59.76/64.70 ≈ 0.9236), but the abstract claims 'near-zero performance loss with 10% visual tokens on general 3D tasks'; please clarify which benchmark and retention ratio each claim refers to.
[§3.4, Eq. (10)] The phrase 'normalize it to match the pruning budget' is ambiguous: Eq. (10) normalizes v to a probability distribution summing to 1, not to the budget K. Please rephrase to avoid confusion, especially since §4.2 later uses group masses of 1/K.
[§5.1] The FST cost weights λ_f, λ_x, λ_τ and capacity weights α_f, α_x, α_τ are all set to 1 in experiments, but no sensitivity analysis is provided. A short robustness study would strengthen the claim that the multi-domain weighting is well-behaved.
[References] Some references are incomplete: [3] lacks a year/venue, and [15] is missing the full proceedings information. Please complete all bibliographic entries.

Circularity Check

0 steps flagged · score 1.0 of 10

No circular derivation: FST cost and capacity are hand-designed, hyperparameters are fixed, and benchmarks are external; the lone self-citation is non-load-bearing.

full rationale

The derivation chain is self-contained in the relevant sense. CoverPrune defines a coverage objective (Eqs. 5-6) from hand-specified FST cost (Eq. 8) and FST capacity (Eq. 10), then approximates it with the SGS algorithm or the block-structured CoverPrune-Lite variant (Eqs. 13-20). The hyperparameters are fixed rather than fitted to test labels: 'We set λf = λx = λτ = 1 and αf = αx = ατ = 1 in our experiments.' The benchmarks (ScanQA, SQA3D, Scan2Cap, VSI-Bench) are external, and the baselines are external methods; the paper's own contributions are evaluated against those benchmarks rather than used as inputs to the derivation. The only self-citation, Ref. 27 (SovGaussian), appears in the Related Work sentence about spatial understanding and plays no role in the pruning formulation, the OT cost design, or the claimed approximation guarantees, so it is not load-bearing and does not constitute circularity. A separate concern raised by a skeptical reader—that under the stated normalization P_l 1 = u and P_l^T 1 ≤ v with equal total mass, the residual in Eq. (14) would be identically zero, making the greedy selection in Eq. (15) degenerate, and that submodularity of the induced set function is asserted without proof—is a correctness/soundness issue rather than a circularity issue. It does not amount to a prediction being equivalent to its inputs by construction. Score 1 reflects only the presence of one non-load-bearing self-citation; the central claim retains independent empirical content.

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

The central claim rests on several hand-set or unspecified hyperparameters and on domain assumptions about the proxy quality of OT coverage, spatial locality, and coordinate estimation. No new physical or model entities are introduced.

free parameters (6)

lambda_f, lambda_x, lambda_tau (FST cost weights) = 1, 1, 1 (stated in Sec 5.1)
Hand-set weights in Eq. (8); no sensitivity analysis or tuning procedure is reported.
alpha_f, alpha_x, alpha_tau (capacity weights) = 1, 1, 1
Hand-set weights in Eq. (9); no sensitivity analysis is reported.
kappa in phi_kappa = not specified
The nonlinear spatial distance scaling in Eq. (8) depends on kappa, but its value is not reported.
neighbor count n in Eq. (9) = not specified
The local distinctiveness score depends on the number of nearest neighbors n, which is not reported.
neighbor count g in Eq. (13) = not specified
The SGS local candidate neighborhood size g is not reported.
capacity mapping phi in Eq. (10) = unspecified monotone increasing mapping
Eq. (10) leaves the mapping phi unspecified, which changes how much capacity concentrates on informative tokens.

assumptions (5)

domain assumption The semi-relaxed OT set function for the FST cost is monotone submodular, so greedy selection has a constant-factor approximation guarantee.
Invoked in Sec 3.5 via refs [17,21]; not proved for the specific FST cost, and the relaxation is tight under the stated normalization.
domain assumption A retained token can effectively cover only spatially nearby target tokens.
Sec 3.5 and Sec 4 use 3D locality to restrict transport; distant but semantically identical observations are assumed non-coverable.
domain assumption OT coverage distortion correlates with downstream spatial reasoning quality.
This is the central premise in Eq. (4)-(6); only benchmark results support it, and no direct validation of the proxy is given.
domain assumption Estimated 3D coordinates from SfM or a geometry foundation model are accurate enough for the spatial cost.
Sec 5.1 says coordinates are generated via an estimator without ground truth; errors propagate into d_x, grouping, and the Morton ordering.
standard math Standard OT and Sinkhorn results apply as used.
Sec 3.1 relies on standard optimal transport definitions and entropic regularization.

how reviews work

0 comments

Cite this review

Pith. "Pith review of CoverPrune: Coverage-Driven Token Pruning for 3D VLMs via Optimal Transport." pith.science (2026). https://pith.science/paper/P3KE55YS

@misc{pith2026260813226,
  author       = {Pith},
  title        = {Pith review of: CoverPrune: Coverage-Driven Token Pruning for 3D VLMs via Optimal Transport},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/P3KE55YS}},
  note         = {Machine review of arXiv:2608.13226}
}

read the original abstract

While 3D Vision-Language Models (3D VLMs) have demonstrated remarkable spatial reasoning capabilities, they suffer from massive visual token counts that create severe computational bottlenecks during inference. Existing token pruning methods primarily rely on diversity-based selection, discarding similar tokens to maximize dispersion. However, in 3D environments, this approach frequently drops representative prototype tokens in favor of outliers, breaking the multi-view consistencies and geometric structures essential for spatial reasoning. In this paper, we propose a paradigm shift for 3D VLM token pruning: from maximizing diversity to preserving visual evidence coverage. We introduce CoverPrune, a training-free framework that formulates inference-time token pruning as an Optimal Transport (OT) problem. To overcome the intractable combinatorial subset selection inherent in this formulation, we design the Feature-Spatial-Temporal (FST) transport cost and target capacity, along with an efficient Spatial-Guided Greedy Selection (SGS) algorithm to approximate the OT objective. Furthermore, we propose CoverPrune-Lite, an accelerated variant utilizing spatially structured local matching for minimal overhead. Extensive experiments across multiple 3D visual-spatial reasoning benchmarks demonstrate that our methods achieve state-of-the-art token efficiency, maintaining robust reasoning performance even under highly aggressive pruning budgets. Visit our project website at https://github.com/Brucess/CoverPrune.

Figures

Figures reproduced from arXiv: 2608.13226 by the authors.

**Figure 1.** CoverPrune and CoverPruneLite Performance. We report crossbenchmark quantitative results under varying token retention ratios, with each dimension representing the average performance retention rate relative to the full token baseline across all metrics for each benchmark; our method achieves near-zero performance loss with 10% visual tokens on general 3D tasks and retains over 90% performance with 15% visual t… view at source ↗

**Figure 2.** Framework overview. (Left) CoverPrune serves as a training-free, plug-andplay module inserted between the visual-geometric encoder and the 3D VLM. (Right) We formulate token pruning as an Optimal Transport (OT) problem to maximize visual evidence coverage. To resolve this, we introduce three key designs: (1) a Feature-SpatialTemporal (FST) Cost C (df , dx, dτ ) to comprehensively model multidimensional token relat… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

42 extracted references · 21 canonical work pages

[1]

Alvar, S.R., Singh, G., Akbari, M., Zhang, Y.: Divprune: Diversity-based visual tokenpruningforlargemultimodalmodels.In:ProceedingsoftheComputerVision and Pattern Recognition Conference. pp. 9392–9401 (2025)

2025
[2]

In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2022)

Azuma, D., Miyanishi, T., Kurita, S., Kawanabe, M.: Scanqa: 3d question answer- ing for spatial scene understanding. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2022)

2022
[3]

In: The Fourteenth International Conference on Learning Representations

Baek, C., Song, J., Kim, S., Kong, K.: An empirical study of attention and di- versity for adaptive visual token pruning in large vision-language models. In: The Fourteenth International Conference on Learning Representations

work page
[4]

arXiv preprint arXiv:2308.129661(2), 3 (2023)

Bai, J., Bai, S., Yang, S., Wang, S., Tan, S., Wang, P., Lin, J., Zhou, C., Zhou, J.: Qwen-vl: A frontier large vision-language model with versatile abilities. arXiv preprint arXiv:2308.129661(2), 3 (2023)

arXiv 2023
[5]

Benamou, J., Carlier, G., Cuturi, M., Nenna, L., Peyré, G.: Iterative bregman projections for regularized transportation problems. SIAM J. Sci. Comput.37(2), A1111–A1138 (2015).https://doi.org/10.1137/141000439,https://doi.org/ 10.1137/141000439

doi:10.1137/141000439 2015
[6]

In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

Cha, J., Kang, W., Mun, J., Roh, B.: Honeybee: Locality-enhanced projector for multimodal llm. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 13817–13827 (2024)

2024
[7]

In: Larochelle, H., Ranzato, M., Hadsell, R., Balcan, M., Lin, H

Chapel, L., Alaya, M.Z., Gasso, G.: Partial optimal transport with applications on positive-unlabeled learning. In: Larochelle, H., Ranzato, M., Hadsell, R., Balcan, M., Lin, H. (eds.) Advances in Neural Information Processing Systems 33: An- nual Conference on Neural Information Processing Systems 2020, NeurIPS 2020, December 6-12, 2020, virtual (2020),h...

work page 2020
[8]

In: European Conference on Computer Vision

Chen, L., Zhao, H., Liu, T., Bai, S., Lin, J., Zhou, C., Chang, B.: An image is worth 1/2 tokens after layer 2: Plug-and-play inference acceleration for large vision-language models. In: European Conference on Computer Vision. pp. 19–35. Springer (2024)

2024

Show all 42 references

[9]

arXiv preprint arXiv:2510.13800 (2025)

Chen, Y., Qi, Z., Zhang, W., Jin, X., Zhang, L., Liu, P.: Reasoning in space via grounding in the world. arXiv preprint arXiv:2510.13800 (2025)

2025
[10]

In: Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition

Chen, Z., Gholami, A., Nießner, M., Chang, A.X.: Scan2cap: Context-aware dense captioning in rgb-d scans. In: Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition. pp. 3193–3203 (2021)

2021
[11]

arXiv preprint arXiv:2509.13317 (2025)

Cheng, A.C., Fu, Y., Chen, Y., Liu, Z., Li, X., Radhakrishnan, S., Han, S., Lu, Y., Kautz, J., Molchanov, P., et al.: 3d aware region prompted vision language model. arXiv preprint arXiv:2509.13317 (2025)

2025
[12]

arXiv preprint arXiv:2507.06261 (2025)

Comanici, G., Bieber, E., Schaekermann, M., Pasupat, I., Sachdeva, N., Dhillon, I., Blistein, M., Ram, O., Zhang, D., Rosen, E., et al.: Gemini 2.5: Pushing the frontier with advanced reasoning, multimodality, long context, and next generation agentic capabilities. arXiv prepr...

2025 arXiv
[13]

Ad- vances in neural information processing systems26(2013)

Cuturi, M.: Sinkhorn distances: Lightspeed computation of optimal transport. Ad- vances in neural information processing systems26(2013)

2013
[14]

arXiv preprint arXiv:2505.20279 (2025)

Fan, Z., Zhang, J., Li, R., Zhang, J., Chen, R., Hu, H., Wang, K., Qu, H., Wang, D., Yan, Z., et al.: Vlm-3r: Vision-language models augmented with instruction-aligned 3d reconstruction. arXiv preprint arXiv:2505.20279 (2025)

2025 arXiv
[15]

In: The Thirty-ninth Annual Conference on Neural Information Processing Systems Datasets and Benchmarks Track

Fu, C., Chen, P., Shen, Y., Qin, Y., Zhang, M., Lin, X., Yang, J., Zheng, X., Li, K., Sun, X., et al.: Mme: A comprehensive evaluation benchmark for multimodal large language models. In: The Thirty-ninth Annual Conference on Neural Information Processing Systems Datasets and B...
[16]

In: European Conference on Computer Vision

Guo, Z., Xu, R., Yao, Y., Cui, J., Ni, Z., Ge, C., Chua, T.S., Liu, Z., Huang, G.: Llava-uhd: an lmm perceiving any aspect ratio and high-resolution images. In: European Conference on Computer Vision. pp. 390–406. Springer (2024)

2024
[17]

In: Dong, Y., Kourtellis, N., Hammer, B., Lozano, J.A

Gurumoorthy, K.S., Jawanpuria, P., Mishra, B.: SPOT: a framework for selection of prototypes using optimal transport. In: Dong, Y., Kourtellis, N., Hammer, B., Lozano, J.A. (eds.) Machine Learning and Knowledge Discovery in Databases. Applied Data Science Track - European Conf...

2021 doi
[18]

In: 2025 IEEE/RSJ International Conference on Intel- ligent Robots and Systems (IROS)

Huang, H.W., Chai, W., Chen, K.M., Yang, C.Y., Hwang, J.N.: Tosa: Token merg- ing with spatial awareness. In: 2025 IEEE/RSJ International Conference on Intel- ligent Robots and Systems (IROS). pp. 9654–9660. IEEE (2025)

2025
[19]

In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

Huang, H.W., Chen, F.C., Chai, W., Su, C.C., Xia, L., Jung, S., Yang, C.Y., Hwang, J.N., Sun, M., Kuo, C.H.: Zero-shot 3d question answering via voxel- based dynamic token compression. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 1...

2025
[20]

In: European conference on computer vision

Huang, K., Zou, H., Xi, Y., Wang, B., Xie, Z., Yu, L.: Ivtp: Instruction-guided visual token pruning for large vision-language models. In: European conference on computer vision. pp. 214–230. Springer (2024)

2024
[21]

Kawano, K., Koide, S., Otaki, K.: Partial wasserstein covering. In: Thirty-Sixth AAAI Conference on Artificial Intelligence, AAAI 2022, Thirty-Fourth Confer- ence on Innovative Applications of Artificial Intelligence, IAAI 2022, The Twel- veth Symposium on Educational Advances...

2022 doi
[22]

Lee, P.Y., Je, J., Park, C., Uy, M.A., Guibas, L., Sung, M.: Perspective-aware reasoninginvision-languagemodelsviamentalimagerysimulation.In:Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 9241–9251 (2025)

2025
[23]

arXiv preprint arXiv:2507.15428 (2025)

Li, J., Li, K., Gao, C., Li, Y., Chen, X.: Egoprune: Efficient token pruning for egomotion video reasoning in embodied agent. arXiv preprint arXiv:2507.15428 (2025)

2025 arXiv
[24]

In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

Li, K., Wang, Y., He, Y., Li, Y., Wang, Y., Liu, Y., Wang, Z., Xu, J., Chen, G., Luo, P., et al.: Mvbench: A comprehensive multi-modal video understanding benchmark. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 22195–22206 (2024)

2024
[25]

International Journal of Computer Vision133(10), 6794–6812 (2025)

Li, W., Yuan, Y., Liu, J., Tang, D., Wang, S., Qin, J., Zhu, J., Zhang, L.: To- kenpacker: Efficient visual projector for multimodal llm. International Journal of Computer Vision133(10), 6794–6812 (2025)

2025
[26]

In: Proceedings of the 2024 conference on empirical methods in natural language processing

Lin, B., Ye, Y., Zhu, B., Cui, J., Ning, M., Jin, P., Yuan, L.: Video-llava: Learning united visual representation by alignment before projection. In: Proceedings of the 2024 conference on empirical methods in natural language processing. pp. 5971– 5984 (2024)

2024
[27]

In: Proceedings of the AAAI Conference on Artificial Intelligence

Ling, P., Tan, T., Lin, J., Yang, W.: Sovgaussian: Sparse-view 3d gaussian splatting for open-vocabulary scene understanding. In: Proceedings of the AAAI Conference on Artificial Intelligence. vol. 39, pp. 5343–5351 (2025)

2025
[28]

Liu, Y., Duan, H., Zhang, Y., Li, B., Zhang, S., Zhao, W., Yuan, Y., Wang, J., He, C., Liu, Z., et al.: Mmbench: Is your multi-modal model an all-around player? In: European conference on computer vision. pp. 216–233. Springer (2024)

2024
[29]

In: International Conference on Learning Representations (2023),https://openreview.net/forum?id=IDJx97BC38

Ma, X., Yong, S., Zheng, Z., Li, Q., Liang, Y., Zhu, S.C., Huang, S.: Sqa3d: Sit- uated question answering in 3d scenes. In: International Conference on Learning Representations (2023),https://openreview.net/forum?id=IDJx97BC38

2023
[30]

Peyré, G., Cuturi, M.: Computational optimal transport. Found. Trends Mach. Learn.11(5-6), 355–607 (2019).https://doi.org/10.1561/2200000073,https: //doi.org/10.1561/2200000073

2019 doi
[31]

Morgan Kaufmann (2006)

Samet, H.: Foundations of multidimensional and metric data structures. Morgan Kaufmann (2006)

2006
[32]

In: The Thirty-ninth Annual Conference on Neural Information Processing Systems

Shen, L., Gong, G., He, T., Zhang, Y., Zhao, S., Ding, G., et al.: Fastvid: Dynamic density pruning for fast video large language models. In: The Thirty-ninth Annual Conference on Neural Information Processing Systems
[33]

Villani, C., et al.: Optimal transport: old and new, vol. 338. Springer (2009)

2009
[34]

arXiv preprint arXiv:2409.12191 (2024)

Wang, P., Bai, S., Tan, S., Wang, S., Fan, Z., Bai, J., Chen, K., Liu, X., Wang, J., Ge, W., et al.: Qwen2-vl: Enhancing vision-language model’s perception of the world at any resolution. arXiv preprint arXiv:2409.12191 (2024)

2024 arXiv
[35]

In: The Thirty-ninth Annual Conference on Neural Information Processing Systems

Wang, Q., Yu, Y., Yuan, Y., Mao, R., Zhou, T.: Videorft: Incentivizing video rea- soning capability in mllms via reinforced fine-tuning. In: The Thirty-ninth Annual Conference on Neural Information Processing Systems
[36]

In: The Thirty-ninth Annual Conference on Neural Information Processing Systems

Wu, D., Liu, F., Hung, Y.H., Duan, Y.: Spatial-mllm: Boosting mllm capabilities in visual-based spatial intelligence. In: The Thirty-ninth Annual Conference on Neural Information Processing Systems
[37]

Advances in Neural Information Pro- cessing Systems37, 28828–28857 (2024) 18 P

Wu, H., Li, D., Chen, B., Li, J.: Longvideobench: A benchmark for long-context interleaved video-language understanding. Advances in Neural Information Pro- cessing Systems37, 28828–28857 (2024) 18 P. Ling et al

2024
[38]

In: Pro- ceedings of the Computer Vision and Pattern Recognition Conference

Yang, J., Yang, S., Gupta, A.W., Han, R., Fei-Fei, L., Xie, S.: Thinking in space: How multimodal large language models see, remember, and recall spaces. In: Pro- ceedings of the Computer Vision and Pattern Recognition Conference. pp. 10632– 10643 (2025)

2025
[39]

In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

Yang, S., Chen, Y., Tian, Z., Wang, C., Li, J., Yu, B., Jia, J.: Visionzip: Longer is better but not necessary in vision language models. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 19792– 19802 (2025)

2025
[40]

In: International Conference on Machine Learning

Zhang, Y., Fan, C.K., Ma, J., Zheng, W., Huang, T., Cheng, K., Gudovskiy, D.A., Okuno, T., Nakata, Y., Keutzer, K., et al.: Sparsevlm: Visual token sparsifica- tion for efficient vision-language model inference. In: International Conference on Machine Learning. pp. 74840–74857...

2025
[41]

In: The Thirty-ninth Annual Conference on Neural Information Processing Systems

Zheng, D., Huang, S., Li, Y., Wang, L.: Learning from videos for 3d world: Enhanc- ing mllms with 3d vision geometry priors. In: The Thirty-ninth Annual Conference on Neural Information Processing Systems
[42]

In: Proceedings of the IEEE/CVF International Conference on Computer Vision

Zhu, Z., Wang, X., Li, Y., Zhang, Z., Ma, X., Chen, Y., Jia, B., Liang, W., Yu, Q., Deng, Z., et al.: Move to understand a 3d scene: Bridging visual grounding and exploration for efficient and versatile embodied navigation. In: Proceedings of the IEEE/CVF International Confere...

2025

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.

[1] [1]

Alvar, S.R., Singh, G., Akbari, M., Zhang, Y.: Divprune: Diversity-based visual tokenpruningforlargemultimodalmodels.In:ProceedingsoftheComputerVision and Pattern Recognition Conference. pp. 9392–9401 (2025)

2025

[2] [2]

In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2022)

Azuma, D., Miyanishi, T., Kurita, S., Kawanabe, M.: Scanqa: 3d question answer- ing for spatial scene understanding. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2022)

2022

[3] [3]

In: The Fourteenth International Conference on Learning Representations

Baek, C., Song, J., Kim, S., Kong, K.: An empirical study of attention and di- versity for adaptive visual token pruning in large vision-language models. In: The Fourteenth International Conference on Learning Representations

work page

[4] [4]

arXiv preprint arXiv:2308.129661(2), 3 (2023)

Bai, J., Bai, S., Yang, S., Wang, S., Tan, S., Wang, P., Lin, J., Zhou, C., Zhou, J.: Qwen-vl: A frontier large vision-language model with versatile abilities. arXiv preprint arXiv:2308.129661(2), 3 (2023)

arXiv 2023

[5] [5]

Benamou, J., Carlier, G., Cuturi, M., Nenna, L., Peyré, G.: Iterative bregman projections for regularized transportation problems. SIAM J. Sci. Comput.37(2), A1111–A1138 (2015).https://doi.org/10.1137/141000439,https://doi.org/ 10.1137/141000439

doi:10.1137/141000439 2015

[6] [6]

In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

Cha, J., Kang, W., Mun, J., Roh, B.: Honeybee: Locality-enhanced projector for multimodal llm. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 13817–13827 (2024)

2024

[7] [7]

In: Larochelle, H., Ranzato, M., Hadsell, R., Balcan, M., Lin, H

Chapel, L., Alaya, M.Z., Gasso, G.: Partial optimal transport with applications on positive-unlabeled learning. In: Larochelle, H., Ranzato, M., Hadsell, R., Balcan, M., Lin, H. (eds.) Advances in Neural Information Processing Systems 33: An- nual Conference on Neural Information Processing Systems 2020, NeurIPS 2020, December 6-12, 2020, virtual (2020),h...

work page 2020

[8] [8]

In: European Conference on Computer Vision

Chen, L., Zhao, H., Liu, T., Bai, S., Lin, J., Zhou, C., Chang, B.: An image is worth 1/2 tokens after layer 2: Plug-and-play inference acceleration for large vision-language models. In: European Conference on Computer Vision. pp. 19–35. Springer (2024)

2024