How Can Driving World Models Do Counterfactual Prediction?

Can Cui; Jiaru Zhang; Ruqi Zhang; Xin Ye; Yi Xu; Ziran Wang

How Can Driving World Models Do Counterfactual Prediction?

T0 review · 0 major / 6 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 Driving world models queried with an alternative action are not performing counterfactual prediction, because they ignore the observed factual continuation; the paper formalizes this gap and shows that supplying the missing evidence…

desk verdict Clean causal diagnosis of a real gap in driving world models, with a careful open-loop benchmark; the central claim holds, limitations are stated, and it deserves refereeing. read the letter →

arxiv 2608.11601 v1 pith:KRUSDZYJ submitted 2026-08-12 cs.CV

Jiaru Zhang , Can Cui , Yi Xu , Xin Ye , Ruqi Zhang , Ziran Wang This is my paper · ORCID

classification cs.CV

keywords counterfactualpredictiondrivingworldmodelscausalinferenceabduction-action-predictionmodelevaluationevidencetransportCARLAbenchmarkautonomous

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

This paper argues that driving world models queried with an alternative action do not deliver counterfactual predictions, because they ignore the factual continuation of the observed episode. The authors formalize the distinction through the causal recipe of abduction, action, and prediction, showing that direct prediction mixes over all worlds consistent with the shared history, while counterfactual prediction conditions on the factual outcome. To make the gap measurable, they build a controlled CARLA benchmark with matched counterfactual ground truth and show that two representative world models score below 0.5 on event recovery, remaining closer to an event-free null than to what actually would have happened. A deliberately simple, training-free pipeline that transports observed evidence into the counterfactual view and lets the frozen model complete the rest raises the recovered fraction to 0.64–0.70 and lowers perceptual distance, supporting the diagnosis.

What carries the argument

The load-bearing object is the pair of world-mixture identities in Eqs. (3) and (4), which separate direct prediction $p(Y|H,a') = \int p(Y|w,a')\,p(w|H)\,dw$ from counterfactual prediction $p(Y_{a'}|H,F^+) = \int p(Y|w,a')\,p(w|H,F^+)\,dw$. This gives the causal analysis its precise content. On the constructive side, the machinery is evidence transport: a frozen depth model lifts each factual continuation frame into a colored 3D point cloud, the known executed and target ego trajectories supply the relative camera pose, and forward splatting with a depth buffer reprojects the points into the counterfactual view to produce a supported evidence region and mask. A frozen world model then completes the unsupported regions, and a Combine stage restores the transported pixels verbatim after completion.

What would settle it

The decisive quantity is the recovered fraction of direct prediction in cases where the factual continuation carries no event information beyond the shared history; the paper's analysis predicts that direct and counterfactual predictions coincide there, so if the recovered fraction stays well below 0.5 or the evidence-transport pipeline still helps, the diagnosis would need revision.

Watch

Extended reading notes

Core claim

The paper's central claim is that direct action-conditioned prediction, $p(Y|H,a')$, is not the counterfactual prediction $p(Y_{a'}|H,F^+)$, because the former weights possible worlds by $p(w|H)$ while the latter weights them by $p(w|H,F^+)$. Whenever the factual continuation carries outcome information absent from the shared history, these two distributions differ. The prescribed computation is the causal recipe of Eq. (2): abduce the realized world from the observed history and continuation, replace the ego action, then predict through the same mechanism. The benchmark experiments confirm the analysis: direct predictions from both a diffusion-based and an autoregressive world model fail to preserve episode-specific events such as a car emerging from a side street, while evidence transport that uses the factual continuation recovers the event.

Load-bearing premise

The load-bearing premise is the open-loop assumption that, over the short prediction window, the alternative ego action changes only the camera viewpoint while surrounding agents follow predetermined behaviors; if those agents would react to the ego's new action within that window, neither the matched replay nor the transported evidence would represent the true counterfactual, a limitation the paper acknowledges in Appendix D.

Editorial extensions

If this is right

Any evaluation of a driving world model as a counterfactual simulator should condition on the factual continuation, not only on the shared history and the alternative action.
Direct predictions can look fluent while depicting the wrong world, so visual quality alone cannot validate counterfactual claim.
Supplying factual evidence through view transport recovers most of the event signal even with frozen model weights, making the gap a missing-input problem rather than a model-capacity problem.
The benchmark's three-arm design, with factual, counterfactual, and null references, offers a template for quantifying counterfactual prediction beyond driving.
Evidence from the wrong time or the wrong episode yields a much lower recovered fraction, showing that the method's success depends on using actual episode information.

Reading between the lines

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

A natural stress test is to extend the protocol to longer horizons with reactive agents; the paper's own limitation section predicts that transported evidence would then preserve behavior the counterfactual action should have changed, so posterior checks that detect and override such evidence are the next step.
Because the pipeline is training-free and every module is replaceable, advances in monocular depth estimation and in world models should transfer directly to this method, possibly closing the residual gap without view-geometry transport.
For planning applications, the argument implies that world-model rollouts that look realistic may still be general futures under an action rather than outcomes tied to the episode being simulated, so counterfactual claims in closed-loop simulation should be treated with caution.
The three-arm benchmark could be adapted to other embodied domains where the same world can be replayed under alternative actions, for example manipulation or pedestrian simulation, to test whether direct action-conditioned generation has the same gap.

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

Colleague,

You should know this paper is a clean, honest piece of work. The central point—that direct action-conditioned prediction p(Y|H,a') is not the same as the counterfactual p(Y_a'|H,F+), because the latter conditions on the observed factual continuation—is exactly right. The authors make it precise with Pearl's abduction-action-prediction recipe, and their Eqs. (3)–(4) state the distinction cleanly. This is not new causal theory, but the targeted application to driving world models is: many papers claim counterfactual capability without noting that they are averaging over worlds consistent with the history, which can wash out episode-specific events.

The strongest part is the benchmark. CARLA lets them replay the same scripted world under the alternative ego action to get a matched counterfactual ground truth, plus an event-free null reference. The three-arm design (factual, counterfactual, null) makes the gap measurable, and the Rec metric is a reasonable preference score. The empirical result is consistent: on both Vista and DrivingWorld, direct predictions score near the null on event recovery, while their training-free transport pipeline recovers most of the lost signal. The ablations and the evidence-source controls (wrong time, wrong episode) are good diagnostics—they show it's the factual evidence that matters, not just pasting extra pixels.

Soft spots, in proportion. The open-loop assumption (surrounding agents follow scripts, so the alternative ego action only changes the camera viewpoint) is the load-bearing scope restriction. The authors know this and say so in Appendix D; over longer horizons, reactive agents would make the transported evidence wrong. That's a genuine limitation but not a flaw in the paper, because the causal distinction does not depend on it. The benchmark is modest (186 cases, 576x320, 10 fps) and both models are out-of-domain (trained on real or other synthetic data, tested on CARLA). The method itself is deliberately simple—it splats depth-based point clouds and lets the world model fill holes—so it's a constructive check, not a final solution. No code or data is released, which is a practical minus for reproducibility, though the protocol is described in enough detail to be reimplemented. None of these undercut the main argument.

Practical take: this is a useful correction to the field's vocabulary. I'd bring it to a reading group and cite it when discussing world models and counterfactual claims. It deserves a serious referee. My recommendation is to accept with minor revisions, mainly asking for code release (or at least the benchmark) and a bit more caution in the abstract about the open-loop scope.

Referee Report

0 major / 6 minor

Summary. The paper examines whether driving world models, when queried with an alternative ego action, perform genuine counterfactual prediction. The authors identify a fundamental mismatch: standard action-conditioned prediction p(Y|H,a') conditions only on the shared history H and the alternative action a', and not on the factual continuation F+ observed after H. They argue that true counterfactual prediction should be p(Y_a'|H,F+), conditioning on the realized episode's outcome. Using Pearl's abduction-action-prediction recipe, they formalize this gap in Eqs. (3)–(4), showing the two predictions differ when F+ carries outcome information absent from H. To make the gap measurable, they build a CARLA benchmark with 186 cases, each providing a factual log, a matched counterfactual ground truth (replay of the same world under the alternative action), and a null reference. Their experiments with two frozen world models (Vista, a diffusion model, and DrivingWorld, an autoregressive model) show that direct predictions consistently fail to match the counterfactual ground truth, typically scoring below 0.5 on a recovered-fraction metric. The authors then propose a simple, training-free pipeline that transports observed evidence from F+ into the counterfactual view and uses the frozen world model to complete unspecified regions. This method raises the recovered fraction to 0.64–0.70 and reduces LPIPS on both backbones.

Significance. The central conceptual contribution is clear and correct: the paper precisely articulates a distinction that is often overlooked when world models are described as counterfactual simulators. The formal difference between p(Y|H,a') and p(Y_a'|H,F+) is cleanly derived and parameter-free, and it provides a principled explanation for why direct action-conditioned prediction can produce fluent but episode-irrelevant futures. The benchmark design, with matched counterfactual replays and a null reference, is a well-controlled instrument that makes the gap quantitatively measurable. The proposed evidence-transport pipeline, though deliberately simple, serves as a compelling constructive check of the diagnosis, and the ablation and evidence-source controls strengthen this interpretation. The paper also explicitly and honestly states its limitations in Appendix D, including the open-loop assumption and the out-of-domain evaluation. These limitations do not undermine the central claim, which is theoretical and independent of the empirical scope. Overall, the paper makes a valuable and timely contribution to the driving world model literature.

minor comments (6)

[3.1 / Appendix D] The open-loop assumption is stated clearly and acknowledged in Appendix D, but its implications for the benchmark's construction should be stated even more prominently in the main text: because surrounding agents are scripted, the event content in F+ is by construction identical to the event content in the counterfactual replay P. This means that the success of Ours is partly baked into the benchmark. The paper's framing as a 'constructive check' mitigates this, but a sentence making the tautology explicit would help readers avoid over-interpreting the empirical gains.
[Appendix D] The out-of-domain evaluation is listed as a limitation, but the main-text discussion of the quantitative results does not revisit this caveat. It would be useful to note, near Table 1, that the absolute recovered-fraction values may be affected by the domain shift between the training data of the world models and the CARLA rendering, even though the comparison between B and Ours on the same frozen backbones remains valid.
[5.1 / Appendix A] The paper does not provide a code or benchmark release statement. Given that the benchmark construction is a significant contribution and the pipeline is training-free, releasing the 186-case benchmark and the evaluation code would substantially improve reproducibility. A data-availability section or statement should be added.
[5.2 / Appendix A.5] The metric definition in Eq. (5) uses s(·,·), while the appendix uses cos(φ(·), φ(·)). The notation should be unified so that readers can follow the exact computation of the recovered fraction without cross-referencing the appendix.
[4.2 / Figure 2] In Figure 2, the header shows '+2.10 m ahead' without explanation. If this refers to a specific scenario parameter or the alternative trajectory offset, it should be defined in the caption or the surrounding text.
[Appendix B] The description of the multi-frame filling threshold and the blurring count is clear, but the values (e.g., the 28-intensity-level threshold and the 60% token coverage threshold) appear to be fixed constants. The paper states that hyperparameters were set during development, which is good; adding a sentence on the sensitivity to these thresholds (or noting that results are stable within a reasonable range) would strengthen the robustness discussion.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the formal distinction in Eqs. (3)–(4) is self-contained and the benchmark is used only for evaluation.

full rationale

The paper's central claim is that direct action-conditioned prediction p(Y|H,a') marginalizes over p(w|H), whereas the counterfactual prediction p(Y_a'|H,F+) uses p(w|H,F+). This follows directly from the standard abduction-action-prediction recipe and is expressed in Eqs. (3)–(4); no parameter is fitted to the counterfactual ground truth, and no load-bearing result is imported from the authors' prior work. The CARLA benchmark supplies P and U only for scoring, and the method's hyperparameters were fixed during development. The proposed pipeline is explicitly a training-free constructive check: it transports the factual continuation into the counterfactual camera view and lets frozen world models complete unsupported regions. Its success on the recovered-fraction metric is therefore partly built into the controlled open-loop benchmark, since the counterfactual ground truth is defined as a replay of the same scripted world under the alternative action. This is an acknowledged limitation (Appendix D), not a circular derivation: the paper does not claim the transported evidence is an independently predicted outcome, and the formal insufficiency of direct prediction is established separately by the conditional-probability argument and by the low scores of the direct baseline. The comparison diagnoses the value of factual evidence rather than fitting the method to the target, so the derivation chain is self-contained.

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

No fundamentally new physical or mathematical entities are introduced. The load-bearing assumptions are the open-loop traffic assumption, the applicability of Pearl's counterfactual recipe, and the reliability of CARLA as a source of ground truth. The free parameters are engineering thresholds in the method, none fitted to the evaluation metric or counterfactual ground truth.

free parameters (4)

Camera height constant for metric depth conversion = 1.8 m
Used in Appendix B to convert relative monocular depth into metric distance before 3D lifting. Set during development and kept fixed; not fitted to the counterfactual ground truth.
Relative depth gradient threshold = 0.15
Pixels at sharp depth changes are removed during transport. Chosen during initial development, not tuned on the evaluation set.
Multi-frame agreement intensity threshold = 28 intensity levels
In the MF filling step, a pixel is filled only when projected colors from multiple frames agree within this spread. Hand-set during development.
DrivingWorld evidence token coverage threshold = 60%
A VQ token is kept fixed when transported evidence covers at least 60% of its image patch. This is a design choice that affects completion behavior.

assumptions (4)

domain assumption Over the prediction window, surrounding agents follow predetermined behaviors and do not react to the alternative ego action.
This open-loop assumption is the basis for the benchmark's matched counterfactual ground truth and for the claim that preserving the factual event is correct. Stated in Section 3.1 and revisited in Appendix D.
domain assumption Pearl's causal hierarchy and the abduction-action-prediction recipe are the correct formalization of counterfactual prediction.
The paper imports Pearl's framework as the standard definition of counterfactuals, using it to distinguish the two sides of Eq. (1). This is a modeling choice, not derived in the paper.
domain assumption A driving world model trained on (history, action, future) triples provides a useful prior p(Y|H,a') for filling regions not determined by transported evidence.
The Complete stage relies on the frozen world model to generate unsupported regions. The paper treats this as a natural use of the model, but does not justify that the prior is appropriate for the CARLA domain.
domain assumption The CARLA simulator replays the same scripted world deterministically enough to serve as exact counterfactual ground truth.
The benchmark is valid only if rerunning the same placement with a different ego action yields a meaningful reference. The paper relies on CARLA's synchronous mode but does not quantify stochasticity or rendering variation.

how reviews work

0 comments

Cite this review

Pith. "Pith review of How Can Driving World Models Do Counterfactual Prediction?." pith.science (2026). https://pith.science/paper/KRUSDZYJ

@misc{pith2026260811601,
  author       = {Pith},
  title        = {Pith review of: How Can Driving World Models Do Counterfactual Prediction?},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KRUSDZYJ}},
  note         = {Machine review of arXiv:2608.11601}
}

read the original abstract

Driving world models are often interpreted as counterfactual simulators for observed driving episodes: given a factual driving log, they are asked what would have happened under an alternative ego action. In this paper, we identify a fundamental mismatch between this goal and direct action-conditioned prediction. The direct prediction uses the shared history and the alternative action but not the factual continuation observed after that history. It can therefore generate a plausible future without preserving what actually happened in this episode. We formalize this gap using the causal recipe of abduction, action, and prediction and study it in a setting with a short time horizon, where the alternative ego action does not alter how surrounding agents evolve. To make the gap measurable, we construct a controlled simulation benchmark with factual outcomes and matched counterfactual outcomes. Across two representative world models, direct predictions fail to match the counterfactual ground truth, supporting our analysis. As a constructive check of this analysis, we introduce a deliberately simple, training-free pipeline that moves observed evidence into the counterfactual view and lets the frozen model complete what remains unknown. Even this simple construction raises the overall recovered fraction substantially and reduces perceptual distance to the matched counterfactual on both models. We hope this work draws attention to this gap and motivates better counterfactual prediction methods for driving world models.

Figures

Figures reproduced from arXiv: 2608.11601 by the authors.

**Figure 2.** Overview of our method. The factual driving log comprises RGB video [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗

**Figure 3.** Qualitative comparison at a representative frame late in the prediction window. The upper [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗

Figures from the paper (1 more)

**Figure 4.** Figure 4: A temporal comparison for a side street case in Town03 with Vista. The columns show [PITH_FULL_IMAGE:figures/full_fig_p016_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

33 extracted references · 25 canonical work pages

[1]

Advances in Neural Information Processing Systems , year =

Gao, Shenyuan and Yang, Jiazhi and Chen, Li and Chitta, Kashyap and Qiu, Yihang and Geiger, Andreas and Zhang, Jun and Li, Hongyang , title =. Advances in Neural Information Processing Systems , year =

work page
[2]

Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , year =

Wang, Yuqi and He, Jiawei and Fan, Lue and Li, Hongxin and Chen, Yuntao and Zhang, Zhaoxiang , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , year =

work page
[3]

2026 , howpublished =

Jiang, Chiyu Max and Masotto, Xander and Sun, Bo , title =. 2026 , howpublished =

work page 2026
[4]

2025 , howpublished =

Parker-Holder, Jack and Fruchter, Shlomi , title =. 2025 , howpublished =

work page 2025
[5]

Pearl, Judea , title =
[6]

European Conference on Computer Vision , year =

Zheng, Wenzhao and Chen, Weiliang and Huang, Yuanhui and Zhang, Borui and Duan, Yueqi and Lu, Jiwen , title =. European Conference on Computer Vision , year =

work page
[7]

arXiv preprint arXiv:2412.19505 , year =

Hu, Xiaotao and Yin, Wei and Jia, Mingkai and Deng, Junyuan and Guo, Xiaoyang and Zhang, Qian and Long, Xiaoxiao and Tan, Ping , title =. arXiv preprint arXiv:2412.19505 , year =

arXiv
[8]

arXiv preprint arXiv:2309.17080 , year =

Hu, Anthony and Russell, Lloyd and Yeo, Hudson and Murez, Zak and Fedoseev, George and Kendall, Alex and Shotton, Jamie and Corrado, Gianluca , title =. arXiv preprint arXiv:2309.17080 , year =

arXiv

Show all 33 references

[9]

European Conference on Computer Vision , year =

Wang, Xiaofeng and Zhu, Zheng and Huang, Guan and Chen, Xinze and Zhu, Jiagang and Lu, Jiwen , title =. European Conference on Computer Vision , year =
[10]

Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , year =

Lugmayr, Andreas and Danelljan, Martin and Romero, Andres and Yu, Fisher and Timofte, Radu and Van Gool, Luc , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , year =
[11]

International Conference on Learning Representations , year =

Meng, Chenlin and He, Yutong and Song, Yang and Song, Jiaming and Wu, Jiajun and Zhu, Jun-Yan and Ermon, Stefano , title =. International Conference on Learning Representations , year =
[12]

Advances in Neural Information Processing Systems , year =

Yang, Lihe and Kang, Bingyi and Huang, Zilong and Zhao, Zhen and Xu, Xiaogang and Feng, Jiashi and Zhao, Hengshuang , title =. Advances in Neural Information Processing Systems , year =
[13]

, title =

Wang, Shihao and Yu, Zhiding and Jiang, Xiaohui and Lan, Shiyi and Shi, Min and Chang, Nadine and Kautz, Jan and Li, Ying and Alvarez, Jose M. , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , year =
[14]

Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages =

Lin, Haohong and Huang, Xin and Phan-Minh, Tung and Hayden, David and Zhang, Huan and Zhao, Ding and Srinivasa, Siddhartha and Wolff, Eric and Chen, Hongge , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages =
[15]

Advances in Neural Information Processing Systems , year =

Yang, Jiazhi and Chitta, Kashyap and Gao, Shenyuan and Chen, Long and Shao, Yuqian and Jia, Xiaosong and Li, Hongyang and Geiger, Andreas and Yue, Xiangyu and Chen, Li , title =. Advances in Neural Information Processing Systems , year =
[16]

Proceedings of the 1st Annual Conference on Robot Learning , year =

Dosovitskiy, Alexey and Ros, German and Codevilla, Felipe and Lopez, Antonio and Koltun, Vladlen , title =. Proceedings of the 1st Annual Conference on Robot Learning , year =
[17]

Transactions on Machine Learning Research , year =

Oquab, Maxime and Darcet, Timoth. Transactions on Machine Learning Research , year =
[18]

International Conference on Machine Learning , year =

Radford, Alec and Kim, Jong Wook and Hallacy, Chris and Ramesh, Aditya and Goh, Gabriel and Agarwal, Sandhini and others , title =. International Conference on Machine Learning , year =
[19]

ACM Transactions on Graphics , volume =

Kerbl, Bernhard and Kopanas, Georgios and Leimk. ACM Transactions on Graphics , volume =
[20]

Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , year =

Shih, Meng-Li and Su, Shih-Yang and Kopf, Johannes and Huang, Jia-Bin , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , year =
[21]

and Shechtman, Eli and Wang, Oliver , title =

Zhang, Richard and Isola, Phillip and Efros, Alexei A. and Shechtman, Eli and Wang, Oliver , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , year =
[22]

European Conference on Computer Vision , year =

Yan, Yunzhi and Lin, Haotong and Zhou, Chenxu and Wang, Weijie and Sun, Haiyang and Zhan, Kun and Lang, Xianpeng and Zhou, Xiaowei and Peng, Sida , title =. European Conference on Computer Vision , year =
[23]

The Essential Role of Causality in Foundation World Models for Embodied

Gupta, Tarun and Gong, Wenbo and Ma, Chao and Pawlowski, Nick and Hilmkil, Agrin and Scetbon, Meyer and Rigter, Marc and Famoti, Ade and Llorens, Ashley Juan and Gao, Jianfeng and Bauer, Stefan and Kragic, Danica and Sch. The Essential Role of Causality in Foundation World Mod...

arXiv
[24]

and Ibeling, Duligur and Icard, Thomas , title =

Bareinboim, Elias and Correa, Juan D. and Ibeling, Duligur and Icard, Thomas , title =. Probabilistic and Causal Inference: The Works of Judea Pearl , publisher =
[25]

Pearl, Judea and Mackenzie, Dana , title =
[26]

, title =

Manski, Charles F. , title =
[27]

Transportation Human Factors , volume =

Green, Marc , title =. Transportation Human Factors , volume =
[28]

Proceedings of the Tenth Conference on Uncertainty in Artificial Intelligence , pages =

Balke, Alexander and Pearl, Judea , title =. Proceedings of the Tenth Conference on Uncertainty in Artificial Intelligence , pages =
[29]

, title =

Holland, Paul W. , title =. Journal of the American Statistical Association , volume =
[30]

Advances in Neural Information Processing Systems , year =

Jia, Xiaosong and Yang, Zhenjie and Li, Qifeng and Zhang, Zhiyuan and Yan, Junchi , title =. Advances in Neural Information Processing Systems , year =
[31]

arXiv preprint arXiv:2106.11810 , year =

Caesar, Holger and Kabzan, Juraj and Tan, Kok Seang and Fong, Whye Kit and Wolff, Eric and Lang, Alex and Fletcher, Luke and Beijbom, Oscar and Omari, Sammy , title =. arXiv preprint arXiv:2106.11810 , year =

arXiv
[32]

European Conference on Computer Vision , year =

Van Hoorick, Basile and Wu, Rundi and Ozguroglu, Ege and Sargent, Kyle and Liu, Ruoshi and Tokmakov, Pavel and Dave, Achal and Zheng, Changxi and Vondrick, Carl , title =. European Conference on Computer Vision , year =
[33]

International Conference on Learning Representations , year =

Wang, Qitai and Fan, Lue and Wang, Yuqi and Chen, Yuntao and Zhang, Zhaoxiang , title =. International Conference on Learning Representations , year =

Pith tools

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

[1] [1]

Advances in Neural Information Processing Systems , year =

Gao, Shenyuan and Yang, Jiazhi and Chen, Li and Chitta, Kashyap and Qiu, Yihang and Geiger, Andreas and Zhang, Jun and Li, Hongyang , title =. Advances in Neural Information Processing Systems , year =

work page

[2] [2]

Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , year =

Wang, Yuqi and He, Jiawei and Fan, Lue and Li, Hongxin and Chen, Yuntao and Zhang, Zhaoxiang , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , year =

work page

[3] [3]

2026 , howpublished =

Jiang, Chiyu Max and Masotto, Xander and Sun, Bo , title =. 2026 , howpublished =

work page 2026

[4] [4]

2025 , howpublished =

Parker-Holder, Jack and Fruchter, Shlomi , title =. 2025 , howpublished =

work page 2025

[5] [5]

Pearl, Judea , title =

[6] [6]

European Conference on Computer Vision , year =

Zheng, Wenzhao and Chen, Weiliang and Huang, Yuanhui and Zhang, Borui and Duan, Yueqi and Lu, Jiwen , title =. European Conference on Computer Vision , year =

work page

[7] [7]

arXiv preprint arXiv:2412.19505 , year =

Hu, Xiaotao and Yin, Wei and Jia, Mingkai and Deng, Junyuan and Guo, Xiaoyang and Zhang, Qian and Long, Xiaoxiao and Tan, Ping , title =. arXiv preprint arXiv:2412.19505 , year =

arXiv

[8] [8]

arXiv preprint arXiv:2309.17080 , year =

Hu, Anthony and Russell, Lloyd and Yeo, Hudson and Murez, Zak and Fedoseev, George and Kendall, Alex and Shotton, Jamie and Corrado, Gianluca , title =. arXiv preprint arXiv:2309.17080 , year =

arXiv