Pith. sign in

REVIEW 4 major objections 4 minor 96 references

Transfer Your Perspective: Controllable 3D Generation from Any Viewpoint in a Driving Scene

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

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

Pith's one-line read The paper claims that a single ego vehicle's LiDAR sweep, together with editable semantic labels, is enough to generate a realistic point cloud from any chosen reference viewpoint in the same scene, and that this generation turns any…

desk verdict A new problem and a useful two-stage pipeline for turning ego-only LiDAR data into collaborative training data; the evaluation needs controls to separate the cross-view transfer from the semantic label prior. read the letter →

arxiv 2502.06682 v2 pith:BHICSO5Z submitted 2025-02-10 cs.CV

Tai-Yu Pan , Sooyoung Jeon , Mengdi Fan , Jinsu Yoo , Zhenyang Feng , Mark Campbell , Kilian Q. Weinberger , Bharath Hariharan
show 1 more author
Wei-Lun Chao
This is my paper · ORCID
classification cs.CV
keywords collaborativeautonomousdrivingLiDARpointcloudgenerationlatentdiffusionmodelsviewpointsynthesisdomainadaptation3Dobjectdetectiondatasetaugmentationbird's-eyeview
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 sets out to remove the main bottleneck in collaborative autonomous driving: the need to collect simultaneous multi-agent sensor data. It claims that a generative model can take one real ego vehicle's LiDAR point cloud and semantic labels, and output a realistic point cloud as seen from any other viewpoint in the same scene, thereby turning ordinary single-agent datasets into collaborative ones. The authors call this problem Transfer Your Perspective (TYP) and propose a two-stage latent diffusion pipeline trained on simulated paired-view data, then adapted to real ego-only data. If the claim holds, collaborative-perception algorithms could be pre-trained at scale without the expensive, coordinated data-collection campaigns that currently limit the field.

What carries the argument

The load-bearing machinery is a latent diffusion model over VQ-VAE tokens of Bird's-Eye-View voxelized LiDAR occupancy. Point clouds are voxelized into $512 \times 512 \times 20$ volumes and encoded into $64 \times 64 \times 8$ feature maps; object boxes become binary BEV maps encoded in the same way. Stage 1 trains a latent diffusion model $P(x|y)$ conditioned only on the object map, using real single-agent data. Stage 2 freezes that model and adds a T2I-Adapter-style control module, a lightweight adapter that injects conditioning features into a frozen diffusion backbone, so that the ego point cloud's encoding enters every scale of the denoising U-Net; this stage is trained on simulated paired views $(x_r,x_e,y_r)$ from OPV2V. Between stages, a discriminator adapts the simulated data's VQ-VAE features to the real target distribution, and optional fine-tuning uses a dummy ego input with self-training on the target dataset.

What would settle it

Take a real scene with a ground-truth reference agent, run TYP with the standard ego labels, and measure whether the generated reference point cloud contains points on objects that are invisible in the ego sweep, such as a pedestrian fully hidden behind a truck. If removing every label of such occluded objects from the reference condition leaves the generated reference view essentially unchanged, then TYP is re-rendering known layouts rather than transferring perspective. The same test can be quantified as downstream detection AP for objects visible only to the reference agent, comparing TYP-generated data against oracle reference data.

Watch

Extended reading notes

Core claim

On its own terms, the central discovery is that viewpoint transfer in driving scenes can be learned without any real paired multi-agent data. TYP trains a conditioned diffusion model in two stages: first it learns to generate realistic ego-style point clouds from object-location maps using real single-agent data; second, with the generator frozen, it learns a lightweight conditioning module that grounds generation on the ego agent's translated point cloud, using simulated collaborative pairs from OPV2V. A GAN-style discriminator aligns simulated and real feature spaces so the resulting model can generate reference-vehicle point clouds for real datasets such as Waymo and V2V4Real. The empirical claim is that detectors trained with these generated point clouds reach performance comparable to using ground-truth reference point clouds (0.46 vs. 0.49 overall AP for early fusion on OPV2V; 0.53 vs. 0.60 on V2V4Real), and that pre-training on the generated 'ColWaymo' version of Waymo improves fine-tuning on V2V4Real beyond pre-training on the simulated OPV2V data.

Load-bearing premise

The method assumes that the labels supplied as the reference-view condition already include objects and surfaces hidden from the ego car, meaning someone must already know about occluded content or manually add it to the labels.

Editorial extensions

If this is right

  • Collaborative perception algorithms trained on TYP-generated reference point clouds reach performance comparable to training on ground-truth reference point clouds, across early, late, and intermediate fusion baselines.
  • A large ego-only dataset like Waymo can be converted into a collaborative dataset, and pre-training on this 'ColWaymo' data improves fine-tuning on real V2V4Real data beyond pre-training on purely simulated OPV2V data.
  • Reference viewpoints are not restricted to actual vehicle positions, so the same model can generate views for arbitrary dynamic agents or static roadside units, and can edit scenes by adding or removing vehicles.
  • The main data-collection bottleneck for CAV shifts from coordinated multi-agent collection to single-agent driving data plus labels, since paired multi-agent data are only needed in simulation.

Reading between the lines

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

  • One implication the paper leaves implicit is that the method's practical value is bounded by the cost of constructing reference-view labels; if occluded objects must be manually added in most scenes, some of the saved collection effort reappears as annotation effort.
  • A direct extension would be to measure downstream collaborative detection AP as a function of how many occluded-object labels are supplied, which would separate the benefit of viewpoint transfer from the benefit of the labels themselves.
  • The same two-stage recipe could in principle be applied to other labeled ego-only datasets, but only where labels cover occluded geometry and static infrastructure, so quantifying label coverage is the natural first test for scaling beyond Waymo.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 4 minor

Summary. The manuscript introduces Transfer Your Perspective (TYP), a method for generating a LiDAR point cloud at a reference viewpoint from an ego agent's point cloud and a semantic map at that viewpoint. The method trains a latent diffusion model in two stages: a first stage learns P(x|y) on real single-agent data, and a second stage, after a domain-adaptation step for the VQ-VAE encoder/decoder from simulated to real data, trains a T2I-Adapter-style control module on simulated paired collaborative data to produce P(xr|xe, yr). The authors validate the approach on OPV2V, apply it as a zero-shot transfer to V2V4Real, and scale it to create ColWaymo from the Waymo Open Dataset. Evaluation is downstream: fusion detectors trained on TYP-generated reference scans are compared with the same detectors trained on ground-truth reference scans, and ColWaymo is used for pre-training before fine-tuning on V2V4Real. The central claim is that TYP can convert a single-agent ego-car dataset into a collaborative-driving dataset, reducing the need for real-world collaborative data collection.

Significance. TYP addresses a real bottleneck: simultaneous multi-agent data collection for collaborative autonomous driving. The manuscript's strengths are the clear problem formulation, the two-stage training recipe with an interleaved domain-adaptation step, and a downstream evaluation that goes beyond generative metrics. The appendix includes repeated-run statistics (Table 8), which is a useful reproducibility practice, and the OPV2V-to-V2V4Real transfer experiment is a meaningful test of generality. If the improvements in Tables 1-3 are attributable to the perspective-transfer mechanism itself, the method would be a valuable data engine for CAV. The current evidence, however, does not isolate that mechanism, because generation is conditioned on object-location labels that may already encode the occluded content, and the experiments never compare against a same-viewpoint control. Resolving this issue is essential before the central claim can be accepted.

major comments (4)
  1. [Sec. 4.3, Tables 1-2] The central comparison between TYP and the oracle is not fully specified. State whether the semantic condition yr used at inference is built from ground-truth object boxes of the scene and whether those same boxes are used as supervision when training the fusion detectors. If both are true, the evaluation is circular in an important sense: the generated xr is conditioned on the annotated object locations, so the reported AP measures whether the generator respects the input boxes rather than whether it discovers occluded geometry. Please report the source of yr in every experiment and add a variant in which yr contains only boxes visible from the ego view.
  2. [Sec. 4.3, Sec. 4.4, Eq. (7)] Add a same-viewpoint control to isolate the contribution of the perspective change. Specifically, run the full pipeline with the reference viewpoint identical to the ego viewpoint (i.e., xe' = xe, with the same yr) and train the fusion detectors on the resulting 'reference' scans. Also report the AP obtained by the stage-1 model P(xr|yr) without the control module. If these controls perform as well as TYP, the gains in Tables 1-3 are due to label-conditioned generation rather than cross-view transfer, which would directly undercut the paper's central claim. If they perform worse, the controls will strengthen the claim.
  3. [Sec. 3.4, Sec. 4.5] The claim of 'little or no real-world collaborative data' depends on the semantic map yr. The paper says object boxes can be 'translated and edited' and that reference locations are 'sampled from labeled vehicles,' but it never quantifies how much of yr consists of content not visible from the ego view. Report, for the ColWaymo generation: (i) the fraction of generated objects in xr that are outside the ego sensor's field of view, (ii) how many boxes were added or edited per scene on average, and (iii) how reference locations were chosen. Without these numbers, the reader cannot assess the data-annotation effort behind the scalability claim.
  4. [Sec. 4.3-4.5, Tables 1-3] Several simple baselines are missing and would help interpret the absolute gains. In particular, compare against (i) using the ego scan xe translated to the reference frame without generation, and (ii) placing a template or box-shaped point pattern at each box in yr. These baselines do not require retraining and would demonstrate whether the learned P(xr|xe, yr) adds value beyond the information already contained in xe and yr. Such controls are inexpensive and would substantially strengthen the claim that the learned cross-view transfer is responsible for the reported improvements.
minor comments (4)
  1. [Sec. 4.3] The parenthetical comparison for AttFuse in the text appears inconsistent with Table 1; the numbers seem reversed relative to the other examples, and the intended order (TYP vs. oracle) should be stated uniformly.
  2. [Sec. 4.3, App. C] There are typos that should be corrected: 'adpot' in the Sec. 4.3 baseline description and 'Fushion' in Table 7 of the appendix.
  3. [Sec. 4.1] The post-processing low/high thresholds and Gaussian perturbation parameters are described only as 'tuned on a small hold-out set'; please report the actual values for reproducibility.
  4. [Table 5] The column label 'Num. Steps' is unclear; it appears to denote the number of training stages rather than sampling steps. Please rename it or clarify in the caption.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the viewpoint-generation model is a conditional generative model trained on paired simulated data and evaluated on external benchmarks, with the semantic-label input explicitly acknowledged.

full rationale

TYP's derivation is self-contained. The generation target is P(xr|xe, yr), where yr is an explicit semantic-map input, not a quantity the paper claims to predict or discover. The paper states in Sec. 3.1: 'This semantic information can be easily translated and edited to become yr ... enabling the reference agent to “see” those objects and surfaces beyond the ego's view' and in Sec. 1: 'object boxes can be manually added to make the map appear more realistic from the reference viewpoint.' Thus the occluded-object content is supplied as conditioning, not derived from xe alone, and the downstream gains on occluded objects are attributable to the input labels as much as to the learned cross-view transfer. This is an experimental attribution gap (a same-viewpoint or label-free control would isolate the contribution), but it is not circular: no equation defines xr in terms of yr by construction, and the generative model is learned from paired (xr, xe, yr) tuples using the diffusion losses in Eqs. (4) and (7), then tested on held-out OPV2V scenes, zero-shot V2V4Real, and fine-tuning on real V2V4Real after ColWaymo pre-training. The evaluation is external: Tables 1-3 compare against ground-truth oracle point clouds and against pre-training on simulated OPV2V, so the central claim is not fitted to the test data. The only author self-citations ([46], [47], [84]) support standard techniques (self-training, a pre-training/fine-tuning protocol, and the challenge of GPS/sync errors) and are not load-bearing for the generative derivation. No uniqueness theorem, ansatz-by-citation, or renamed known result appears. The paper's own limitation statement (Sec. 5: 'TYP focuses on vehicle-like objects') narrows scope but does not indicate circularity. Overall, no circular step is present.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The central claim rests on the availability of scene-level semantic labels, on sim-to-real transfer of the viewpoint mapping, and on the BEV occupancy representation being good enough for downstream detection. The only explicit fitted constants are standard hyperparameters and post-processing thresholds; no new physical entities are introduced.

free parameters (2)
  • Voxel grid range and resolution = [-51.2, 51.2] m x/y, 4 m z, 512x512x20 (dataset-dependent)
    Sec 4.1; this discretization controls how much spatial detail the generator can represent and is chosen by hand per dataset.
  • Post-processing low/high thresholds and Gaussian perturbation = not disclosed (tuned on a small hold-out set)
    Sec 4.1; the thresholds convert occupancy logits to binary voxels and can influence downstream detection AP; their values are not reported.
assumptions (4)
  • ad hoc to paper Scene-level semantic labels yr covering objects beyond the ego field of view are available or can be manually added.
    Sec 3.1: 'This semantic information can be easily translated and edited to become yr... enabling the reference agent to see those objects and surfaces beyond the ego's view.' This is essential for the claim of generating occluded content.
  • domain assumption The ego-to-reference viewpoint mapping learned from the CARLA-based OPV2V simulator transfers to real LiDAR sensors.
    Sec 4.4 applies an OPV2V-trained model to V2V4Real without adaptation; the modest performance gaps in Table 2 depend on this transfer holding.
  • domain assumption BEV voxel occupancy at 512x512x20 with a 51.2m range is a faithful representation of LiDAR for training downstream detectors.
    Sec 3.2 and Sec 4.1; the entire generation pipeline and the downstream PointPillars training operate on this discretized representation.
  • domain assumption The first-stage prior P(x|y) trained on real single-agent data provides a sufficient target-domain prior for the second-stage grounding.
    Sec 3.3 Stage 1; if the real single-agent dataset is not representative of the target domain, the generated reference views inherit that mismatch.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Transfer Your Perspective: Controllable 3D Generation from Any Viewpoint in a Driving Scene." pith.science (2026). https://pith.science/paper/BHICSO5Z

@misc{pith2026250206682,
  author       = {Pith},
  title        = {Pith review of: Transfer Your Perspective: Controllable 3D Generation from Any Viewpoint in a Driving Scene},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BHICSO5Z}},
  note         = {Machine review of arXiv:2502.06682}
}
read the original abstract

Self-driving cars relying solely on ego-centric perception face limitations in sensing, often failing to detect occluded, faraway objects. Collaborative autonomous driving (CAV) seems like a promising direction, but collecting data for development is non-trivial. It requires placing multiple sensor-equipped agents in a real-world driving scene, simultaneously! As such, existing datasets are limited in locations and agents. We introduce a novel surrogate to the rescue, which is to generate realistic perception from different viewpoints in a driving scene, conditioned on a real-world sample - the ego-car's sensory data. This surrogate has huge potential: it could potentially turn any ego-car dataset into a collaborative driving one to scale up the development of CAV. We present the very first solution, using a combination of simulated collaborative data and real ego-car data. Our method, Transfer Your Perspective (TYP), learns a conditioned diffusion model whose output samples are not only realistic but also consistent in both semantics and layouts with the given ego-car data. Empirical results demonstrate TYP's effectiveness in aiding in a CAV setting. In particular, TYP enables us to (pre-)train collaborative perception algorithms like early and late fusion with little or no real-world collaborative data, greatly facilitating downstream CAV applications.

Figures

Figures reproduced from arXiv: 2502.06682 by the authors.

Figure 1
Figure 1. Illustration of the proposed problem and solution, Transfer Your Perspective (TYP). (a) A given sensory data captured by the ego-car (red triangle). (b) A generated sensory data by TYP, seeing from the viewpoint of another vehicle (green triangle) in the same scene. (c) A generated sensory data, seeing from an imaginary static agent like roadside units (blue icon). (d) Putting all the sensory data together, given or… view at source ↗
Figure 2
Figure 2. Illustration of TYP’s conditioned generative model and training process. We propose a two-stage training procedure. The first stage maximizes the generation capability by conditioning solely on object locations (using real single-agent target data), while the second stage grounds the generation on the ego-car’s perspective to match semantics and layouts (using simulated CAV data). Additionally, we introduce a discri… view at source ↗
Figure 3
Figure 3. Qualitative results on enhancement in the target domain. Generated point cloud (green) has better quality with the enhancement given ego (gray) from Waymo (cf . Sec. 3.4). Recall that in training P(xr|xe, yr) using the source CAV data, point clouds show up in both the input as con￾ditions and the outputs as supervisions. In other words, the domain gap exists on both sides, and we aim to reduce it by further fine-tun… view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Qualitative results on Collaborative Waymo. The gray point clouds are from the original single-agent dataset and the green are generated by TYP conditioning on them. car removed car added normal-traffic day light-traffic day heavy-traffic day ego-centric car (a) (b) (c…
Figure 5
Figure 5. Figure 5: Qualitative results. Our proposed TYP is capable of scene editing, by inputting the same point cloud but different object locations. We (b) remove and (c) add a car from (a) the original point cloud. Inspired by the idea of past traversals [85], we apply completely dif…
Figure 6
Figure 6. Figure 6: Visualization with validation data of OPV2V. The gen￾erated point clouds are well-aligned with the ground-truth bounding boxes and follow the physics (e.g., occluded areas). converted into multi-agent data for autonomous driving. B. Existing Datasets for CAV We summari…
Figure 7
Figure 7. Figure 7: Illustration of the proposed problem and solution, Transfer Your Perspective (TYP). (a) A given sensory data captured by the ego-car (red triangle). (b) A generated sensory data by TYP, seeing from the viewpoint of another vehicle (green triangle) in the same scene. (c…
Figure 8
Figure 8. Figure 8: Qualitative results on Collaborative Waymo. The gray point clouds are from the original single-agent dataset and the green are generated by TYP conditioning on them. 15 [PITH_FULL_IMAGE:figures/full_fig_p015_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

96 extracted references · 71 canonical work pages

  1. [1]

    Learning representations and generative models for 3d point clouds

    Panos Achlioptas, Olga Diamanti, Ioannis Mitliagkas, and Leonidas Guibas. Learning representations and generative models for 3d point clouds. In ICML, 2018. 12

    2018

  2. [2]

    Domain-adversarial neural networks

    Hana Ajakan, Pascal Germain, Hugo Larochelle, Fran c ¸ois Laviolette, and Mario Marchand. Domain-adversarial neural networks. arXiv preprint arXiv:1412.4446, 2014. 13

    arXiv 2014

  3. [3]

    ediff-i: Text-to-image diffusion models with an ensemble of expert denoisers

    Yogesh Balaji, Seungjun Nah, Xun Huang, Arash Vahdat, Jiaming Song, Qinsheng Zhang, Karsten Kreis, Miika Ait- tala, Timo Aila, Samuli Laine, et al. ediff-i: Text-to-image diffusion models with an ensemble of expert denoisers. arXiv preprint arXiv:2211.01324, 2022. 12

    arXiv 2022

  4. [4]

    Mip-nerf: A multiscale representation for anti-aliasing neural radiance fields

    Jonathan T Barron, Ben Mildenhall, Matthew Tancik, Peter Hedman, Ricardo Martin-Brualla, and Pratul P Srinivasan. Mip-nerf: A multiscale representation for anti-aliasing neural radiance fields. In CVPR, 2021. 13

    2021

  5. [5]

    Mip-nerf 360: Unbounded anti-aliased neural radiance fields

    Jonathan T Barron, Ben Mildenhall, Dor Verbin, Pratul P Srinivasan, and Peter Hedman. Mip-nerf 360: Unbounded anti-aliased neural radiance fields. In CVPR, 2022. 13

    2022

  6. [6]

    Also: Automotive lidar self- supervision by occupancy estimation

    Alexandre Boulch, Corentin Sautier, Bj¨orn Michele, Gilles Puy, and Renaud Marlet. Also: Automotive lidar self- supervision by occupancy estimation. In CVPR, 2023. 8

    2023

  7. [7]

    nuscenes: A multimodal 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 multimodal dataset for autonomous driving. In CVPR, 2020. 4, 5

    2020

  8. [8]

    F-cooper: Feature based cooperative perception for autonomous vehicle edge computing system using 3d point clouds

    Qi Chen, Xu Ma, Sihai Tang, Jingda Guo, Qing Yang, and Song Fu. F-cooper: Feature based cooperative perception for autonomous vehicle edge computing system using 3d point clouds. In Proceedings of the 4th ACM/IEEE Symposium on Edge Computing, 2019. 1

    2019

Show all 96 references
  1. [9]

    Cooper: Cooperative perception for connected autonomous vehicles based on 3d point clouds

    Qi Chen, Sihai Tang, Qing Yang, and Song Fu. Cooper: Cooperative perception for connected autonomous vehicles based on 3d point clouds. In International Conference on Distributed Computing Systems (ICDCS). IEEE, 2019. 1, 6, 7, 13, 14

    2019

  2. [10]

    Multi-view 3d object detection network for autonomous driv- ing

    Xiaozhi Chen, Huimin Ma, Ji Wan, Bo Li, and Tian Xia. Multi-view 3d object detection network for autonomous driv- ing. In CVPR, 2017. 4

    2017

  3. [11]

    Depth-supervised nerf: Fewer views and faster training for free

    Kangle Deng, Andrew Liu, Jun-Yan Zhu, and Deva Ramanan. Depth-supervised nerf: Fewer views and faster training for free. In CVPR, 2022. 13

    2022

  4. [12]

    Diffusion models beat gans on image synthesis

    Prafulla Dhariwal and Alexander Nichol. Diffusion models beat gans on image synthesis. In NeurIPS, 2021. 4, 12

    2021

  5. [13]

    Cogview: Mastering text-to-image generation via transformers

    Ming Ding, Zhuoyi Yang, Wenyi Hong, Wendi Zheng, Chang Zhou, Da Yin, Junyang Lin, Xu Zou, Zhou Shao, Hongxia Yang, et al. Cogview: Mastering text-to-image generation via transformers. In NeurIPS, 2021. 12

    2021

  6. [14]

    CARLA: An open urban driving simulator

    Alexey Dosovitskiy, German Ros, Felipe Codevilla, Antonio Lopez, and Vladlen Koltun. CARLA: An open urban driving simulator. In CoRL, 2017. 2, 3

    2017

  7. [15]

    A point set generation network for 3d object reconstruction from a single image

    Haoqiang Fan, Hao Su, and Leonidas J Guibas. A point set generation network for 3d object reconstruction from a single image. In CVPR, 2017. 12

    2017

  8. [16]

    Training-free structured diffusion guidance for compositional text-to-image synthesis

    Weixi Feng, Xuehai He, Tsu-Jui Fu, Varun Jampani, Arjun Akula, Pradyumna Narayana, Sugato Basu, Xin Eric Wang, and William Yang Wang. Training-free structured diffusion guidance for compositional text-to-image synthesis. In ICLR,

  9. [17]

    Make-a-scene: Scene-based text-to-image generation with human priors

    Oran Gafni, Adam Polyak, Oron Ashual, Shelly Sheynin, Devi Parikh, and Yaniv Taigman. Make-a-scene: Scene-based text-to-image generation with human priors. In ECCV, 2022. 12

    2022

  10. [18]

    Unsupervised domain adaptation by backpropagation

    Yaroslav Ganin and Victor Lempitsky. Unsupervised domain adaptation by backpropagation. In ICML, 2015. 13

    2015

  11. [19]

    Domain-adversarial training of neural networks

    Yaroslav Ganin, Evgeniya Ustinova, Hana Ajakan, Pascal Germain, Hugo Larochelle, Franc ¸ois Laviolette, Mario March, and Victor Lempitsky. Domain-adversarial training of neural networks. Journal of machine learning research, 17(59):1–35,

  12. [20]

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

    Andreas Geiger, Philip Lenz, and Raquel Urtasun. Are we ready for autonomous driving? the kitti vision benchmark suite. In CVPR, 2012. 4

    2012

  13. [21]

    Generative adversarial nets

    Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial nets. In NeurIPS,

  14. [22]

    Waymax: An accelerated, data-driven simulator for large-scale autonomous driving research

    Cole Gulino, Justin Fu, Wenjie Luo, George Tucker, Eli Bron- stein, Yiren Lu, Jean Harb, Xinlei Pan, Yan Wang, Xiangyu Chen, et al. Waymax: An accelerated, data-driven simulator for large-scale autonomous driving research. In NeurIPS,

  15. [23]

    Prompt-to-prompt image editing with cross-attention control

    Amir Hertz, Ron Mokady, Jay Tenenbaum, Kfir Aberman, Yael Pritch, and Daniel Cohen-or. Prompt-to-prompt image editing with cross-attention control. In ICLR, 2023. 12

    2023

  16. [24]

    Denoising diffu- sion probabilistic models

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffu- sion probabilistic models. In NeurIPS, 2020. 3, 4, 12

    2020

  17. [25]

    Rangeldm: Fast realistic lidar point cloud generation

    Qianjiang Hu, Zhimin Zhang, and Wei Hu. Rangeldm: Fast realistic lidar point cloud generation. In ECCV, 2025. 2, 3, 12

    2025

  18. [26]

    Where2comm: Communication-efficient collaborative perception via spatial confidence maps

    Yue Hu, Shaoheng Fang, Zixing Lei, Yiqi Zhong, and Siheng Chen. Where2comm: Communication-efficient collaborative perception via spatial confidence maps. In NeurIPS, 2022. 1

    2022

  19. [27]

    Neural lidar fields for novel view synthesis

    Shengyu Huang, Zan Gojcic, Zian Wang, Francis Williams, Yoni Kasten, Sanja Fidler, Konrad Schindler, and Or Litany. Neural lidar fields for novel view synthesis. In ICCV, 2023. 13

    2023

  20. [28]

    Image-to-image translation with conditional adversarial net- works

    Phillip Isola, Jun-Yan Zhu, Tinghui Zhou, and Alexei A Efros. Image-to-image translation with conditional adversarial net- works. In CVPR, 2017. 13

    2017

  21. [29]

    Categorical reparam- eterization with gumbel-softmax

    Eric Jang, Shixiang Gu, and Ben Poole. Categorical reparam- eterization with gumbel-softmax. In ICLR, 2017. 6

    2017

  22. [30]

    Variational diffusion models

    Diederik Kingma, Tim Salimans, Ben Poole, and Jonathan Ho. Variational diffusion models. In NeurIPS, 2021. 4, 12 9

    2021

  23. [31]

    Pointpillars: Fast encoders for object detection from point clouds

    Alex H Lang, Sourabh V ora, Holger Caesar, Lubing Zhou, Jiong Yang, and Oscar Beijbom. Pointpillars: Fast encoders for object detection from point clouds. In CVPR, 2019. 4, 6

    2019

  24. [32]

    Pseudo-label: The simple and efficient semi-supervised learning method for deep neural networks

    Dong-Hyun Lee et al. Pseudo-label: The simple and efficient semi-supervised learning method for deep neural networks. In Workshop on challenges in representation learning, ICML,

  25. [33]

    Controlnet++: Improv- ing conditional controls with efficient consistency feedback

    Ming Li, Taojiannan Yang, Huafeng Kuang, Jie Wu, Zhaoning Wang, Xuefeng Xiao, and Chen Chen. Controlnet++: Improv- ing conditional controls with efficient consistency feedback. In ECCV, 2025. 12

    2025

  26. [34]

    Learning distilled collaboration graph for multi-agent perception

    Yiming Li, Shunli Ren, Pengxiang Wu, Siheng Chen, Chen Feng, and Wenjun Zhang. Learning distilled collaboration graph for multi-agent perception. In NeurIPS, 2021. 1

    2021

  27. [35]

    V2x-sim: Multi-agent collab- orative perception dataset and benchmark for autonomous driving

    Yiming Li, Dekun Ma, Ziyan An, Zixun Wang, Yiqi Zhong, Siheng Chen, and Chen Feng. V2x-sim: Multi-agent collab- orative perception dataset and benchmark for autonomous driving. In Robotics and Automation Letters, pages 10914– 10921. IEEE, 2022. 3

    2022

  28. [36]

    Kitti-360: A novel dataset and benchmarks for urban scene understanding in 2d and 3d

    Yiyi Liao, Jun Xie, and Andreas Geiger. Kitti-360: A novel dataset and benchmarks for urban scene understanding in 2d and 3d. IEEE TPAMI, 45(3):3292–3310, 2022. 5

    2022

  29. [37]

    Geometric gan

    Jae Hyun Lim and Jong Chul Ye. Geometric gan. arXiv preprint arXiv:1705.02894, 2017. 5, 13

    2017 arXiv

  30. [38]

    When2com: Multi-agent perception via communication graph grouping

    Yen-Cheng Liu, Junjiao Tian, Nathaniel Glaser, and Zsolt Kira. When2com: Multi-agent perception via communication graph grouping. In CVPR, 2020. 1

    2020

  31. [39]

    Learning transferable features with deep adaptation networks

    Mingsheng Long, Yue Cao, Jianmin Wang, and Michael Jor- dan. Learning transferable features with deep adaptation networks. In ICML, 2015. 12

    2015

  32. [40]

    V oxnet: A 3d convo- lutional neural network for real-time object recognition

    Daniel Maturana and Sebastian Scherer. V oxnet: A 3d convo- lutional neural network for real-time object recognition. In IROS, 2015. 4

    2015

  33. [41]

    Nerf: Representing scenes as neural radiance fields for view synthe- sis

    Ben Mildenhall, Pratul P Srinivasan, Matthew Tancik, Jonathan T Barron, Ravi Ramamoorthi, and Ren Ng. Nerf: Representing scenes as neural radiance fields for view synthe- sis. In ECCV. ACM New York, NY , USA, 2021. 13

    2021

  34. [42]

    Rangenet++: Fast and accurate lidar semantic segmenta- tion

    Andres Milioto, Ignacio Vizzo, Jens Behley, and Cyrill Stach- niss. Rangenet++: Fast and accurate lidar semantic segmenta- tion. In IROS, 2019. 4

    2019

  35. [43]

    T2i-adapter: Learning adapters to dig out more controllable ability for text-to-image diffusion models

    Chong Mou, Xintao Wang, Liangbin Xie, Yanze Wu, Jian Zhang, Zhongang Qi, and Ying Shan. T2i-adapter: Learning adapters to dig out more controllable ability for text-to-image diffusion models. In AAAI, 2024. 5, 6, 12

    2024

  36. [44]

    Lidar data synthesis with denoising diffusion probabilistic models

    Kazuto Nakashima and Ryo Kurazume. Lidar data synthesis with denoising diffusion probabilistic models. In ICRA, 2024. 2, 3, 12

    2024

  37. [45]

    Glide: Towards photorealistic image genera- tion and editing with text-guided diffusion models

    Alexander Quinn Nichol, Prafulla Dhariwal, Aditya Ramesh, Pranav Shyam, Pamela Mishkin, Bob Mcgrew, Ilya Sutskever, and Mark Chen. Glide: Towards photorealistic image genera- tion and editing with text-guided diffusion models. In ICML,

  38. [46]

    To- wards open-world segmentation of parts

    Tai-Yu Pan, Qing Liu, Wei-Lun Chao, and Brian Price. To- wards open-world segmentation of parts. In CVPR, 2023. 6

    2023

  39. [47]

    Pre-training lidar-based 3d object detectors through colorization

    Tai-Yu Pan, Chenyang Ma, Tianle Chen, Cheng Perng Phoo, Katie Z Luo, Yurong You, Mark Campbell, Kilian Q Wein- berger, Bharath Hariharan, and Wei-Lun Chao. Pre-training lidar-based 3d object detectors through colorization. In ICLR,

  40. [48]

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

    Charles R Qi, Hao Su, Kaichun Mo, and Leonidas J Guibas. Pointnet: Deep learning on point sets for 3d classification and segmentation. In CVPR, 2017. 4

    2017

  41. [49]

    Pointnet++: Deep hierarchical feature learning on point sets in a metric space

    Charles Ruizhongtai Qi, Li Yi, Hao Su, and Leonidas J Guibas. Pointnet++: Deep hierarchical feature learning on point sets in a metric space. In NeurIPS, 2017. 4

    2017

  42. [50]

    Zero-shot text-to-image generation

    Aditya Ramesh, Mikhail Pavlov, Gabriel Goh, Scott Gray, Chelsea V oss, Alec Radford, Mark Chen, and Ilya Sutskever. Zero-shot text-to-image generation. In ICML, 2021. 12

    2021

  43. [51]

    Towards realistic scene generation with lidar diffusion models

    Haoxi Ran, Vitor Guizilini, and Yue Wang. Towards realistic scene generation with lidar diffusion models. In CVPR, 2024. 2, 3, 12

    2024

  44. [52]

    Collabora- tive automated driving: A machine learning-based method to enhance the accuracy of shared information

    Zaydoun Yahya Rawashdeh and Zheng Wang. Collabora- tive automated driving: A machine learning-based method to enhance the accuracy of shared information. In Inter- national Conference on Intelligent Transportation Systems (ITSC). IEEE, 2018. 1

    2018

  45. [53]

    Dense depth priors for neural radiance fields from sparse input views

    Barbara Roessle, Jonathan T Barron, Ben Mildenhall, Pratul P Srinivasan, and Matthias Nießner. Dense depth priors for neural radiance fields from sparse input views. In CVPR,

  46. [54]

    High-resolution image synthesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bj ¨orn Ommer. High-resolution image synthesis with latent diffusion models. In CVPR, 2022. 2, 4, 12

    2022

  47. [55]

    U-net: Convolutional networks for biomedical image segmentation

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net: Convolutional networks for biomedical image segmentation. In MICCAI, 2015. 5

    2015

  48. [56]

    Focal loss for dense object detection

    T-YLPG Ross and GKHP Doll´ar. Focal loss for dense object detection. In CVPR, 2017. 4

    2017

  49. [57]

    Pho- torealistic text-to-image diffusion models with deep language understanding

    Chitwan Saharia, William Chan, Saurabh Saxena, Lala Li, Jay Whang, Emily L Denton, Kamyar Ghasemipour, Raphael Gontijo Lopes, Burcu Karagol Ayan, Tim Salimans, et al. Pho- torealistic text-to-image diffusion models with deep language understanding. In NeurIPS, 2022. 12

    2022

  50. [58]

    Pointr- cnn: 3d object proposal generation and detection from point cloud

    Shaoshuai Shi, Xiaogang Wang, and Hongsheng Li. Pointr- cnn: 3d object proposal generation and detection from point cloud. In CVPR, 2019. 4

    2019

  51. [59]

    3d point cloud generative adversarial network based on tree struc- tured graph convolutions

    Dong Wook Shu, Sung Woo Park, and Junseok Kwon. 3d point cloud generative adversarial network based on tree struc- tured graph convolutions. In ICCV, 2019. 12

    2019

  52. [60]

    Deep unsupervised learning using nonequilibrium thermodynamics

    Jascha Sohl-Dickstein, Eric Weiss, Niru Maheswaranathan, and Surya Ganguli. Deep unsupervised learning using nonequilibrium thermodynamics. In ICML, 2015. 12

    2015

  53. [61]

    Denoising diffusion implicit models

    Jiaming Song, Chenlin Meng, and Stefano Ermon. Denoising diffusion implicit models. In ICLR, 2021. 6

    2021

  54. [62]

    Deep coral: Correlation alignment for deep domain adaptation

    Baochen Sun and Kate Saenko. Deep coral: Correlation alignment for deep domain adaptation. In ECCV Workshops,

  55. [63]

    Scalability in perception for autonomous driving: Waymo open dataset

    Pei Sun, Henrik Kretzschmar, Xerxes Dotiwalla, Aurelien Chouard, Vijaysai Patnaik, Paul Tsui, James Guo, Yin Zhou, Yuning Chai, Benjamin Caine, et al. Scalability in perception for autonomous driving: Waymo open dataset. InCVPR, 2020. 2, 3, 4, 5, 6, 7, 8 10

    2020

  56. [64]

    Lidar-nerf: Novel lidar view synthesis via neural radiance fields

    Tang Tao, Longfei Gao, Guangrun Wang, Yixing Lao, Peng Chen, Hengshuang Zhao, Dayang Hao, Xiaodan Liang, Math- ieu Salzmann, and Kaicheng Yu. Lidar-nerf: Novel lidar view synthesis via neural radiance fields. In ACM MM, 2024. 13

    2024

  57. [65]

    Deep domain confusion: Maximizing for domain invariance

    Eric Tzeng, Judy Hoffman, Ning Zhang, Kate Saenko, and Trevor Darrell. Deep domain confusion: Maximizing for domain invariance. arXiv preprint arXiv:1412.3474, 2014. 12

    2014 arXiv

  58. [66]

    Simultaneous deep transfer across domains and tasks

    Eric Tzeng, Judy Hoffman, Trevor Darrell, and Kate Saenko. Simultaneous deep transfer across domains and tasks. In ICCV, 2015. 12

    2015

  59. [67]

    Adversarial discriminative domain adaptation

    Eric Tzeng, Judy Hoffman, Kate Saenko, and Trevor Darrell. Adversarial discriminative domain adaptation. InCVPR, 2017. 2, 5, 13

    2017

  60. [68]

    Learn- ing localized generative models for 3d point clouds via graph convolution

    Diego Valsesia, Giulia Fracastoro, and Enrico Magli. Learn- ing localized generative models for 3d point clouds via graph convolution. In ICLR, 2018. 12

    2018

  61. [69]

    Neural discrete representation learning

    Aaron Van Den Oord, Oriol Vinyals, et al. Neural discrete representation learning. In NeurIPS, 2017. 4, 6

    2017

  62. [70]

    Pretraining is all you need for image-to-image translation

    Tengfei Wang, Ting Zhang, Bo Zhang, Hao Ouyang, Dong Chen, Qifeng Chen, and Fang Wen. Pretraining is all you need for image-to-image translation. arXiv preprint arXiv:2205.12952, 2022. 12

    2022 arXiv

  63. [71]

    V2vnet: Vehicle- to-vehicle communication for joint perception and prediction

    Tsun-Hsuan Wang, Sivabalan Manivasagam, Ming Liang, Bin Yang, Wenyuan Zeng, and Raquel Urtasun. V2vnet: Vehicle- to-vehicle communication for joint perception and prediction. In ECCV, 2020. 1

    2020

  64. [72]

    Squeezeseg: Convolutional neural nets with recurrent crf for real-time road-object segmentation from 3d lidar point cloud

    Bichen Wu, Alvin Wan, Xiangyu Yue, and Kurt Keutzer. Squeezeseg: Convolutional neural nets with recurrent crf for real-time road-object segmentation from 3d lidar point cloud. In ICRA, 2018. 4

    2018

  65. [73]

    Text2lidar: Text-guided lidar point cloud generation via equirectangular transformer

    Yang Wu, Kaihua Zhang, Jianjun Qian, Jin Xie, and Jian Yang. Text2lidar: Text-guided lidar point cloud generation via equirectangular transformer. In ECCV. Springer, 2024. 2, 3, 12

    2024

  66. [74]

    V2x-real: a largs-scale dataset for vehicle-to-everything cooperative perception

    Hao Xiang, Zhaoliang Zheng, Xin Xia, Runsheng Xu, Letian Gao, Zewei Zhou, Xu Han, Xinkai Ji, Mingxi Li, Zonglin Meng, et al. V2x-real: a largs-scale dataset for vehicle-to-everything cooperative perception. arXiv preprint arXiv:2403.16034, 2024. 1

    2024 arXiv

  67. [75]

    Pandaset: Advanced sensor suite dataset for au- tonomous driving

    Pengchuan Xiao, Zhenlei Shao, Steven Hao, Zishuo Zhang, Xiaolin Chai, Judy Jiao, Zesong Li, Jian Wu, Kai Sun, Kun Jiang, et al. Pandaset: Advanced sensor suite dataset for au- tonomous driving. In International Intelligent Transportation Systems Conference (ITSC). IEEE, 2021. 5

    2021

  68. [76]

    Pointcontrast: Unsupervised pre- training for 3d point cloud understanding

    Saining Xie, Jiatao Gu, Demi Guo, Charles R Qi, Leonidas Guibas, and Or Litany. Pointcontrast: Unsupervised pre- training for 3d point cloud understanding. In ECCV, 2020. 8

    2020

  69. [77]

    Learning compact representations for lidar completion and generation

    Yuwen Xiong, Wei-Chiu Ma, Jingkang Wang, and Raquel Ur- tasun. Learning compact representations for lidar completion and generation. In CVPR, 2023. 3, 4

    2023

  70. [78]

    Ultralidar: Learning compact representations for lidar completion and generation

    Yuwen Xiong, Wei-Chiu Ma, Jingkang Wang, and Raquel Urtasun. Ultralidar: Learning compact representations for lidar completion and generation. In CVPR, 2023. 2, 4, 12

    2023

  71. [79]

    V2x-vit: Vehicle-to-everything cooperative perception with vision transformer

    Runsheng Xu, Hao Xiang, Zhengzhong Tu, Xin Xia, Ming- Hsuan Yang, and Jiaqi Ma. V2x-vit: Vehicle-to-everything cooperative perception with vision transformer. In ECCV,

  72. [80]

    Opv2v: An open benchmark dataset and fusion pipeline for perception with vehicle-to-vehicle communica- tion

    Runsheng Xu, Hao Xiang, Xin Xia, Xu Han, Jinlong Li, and Jiaqi Ma. Opv2v: An open benchmark dataset and fusion pipeline for perception with vehicle-to-vehicle communica- tion. In ICRA, 2022. 2, 3, 5, 6, 7, 8, 13, 14

    2022

  73. [81]

    V2v4real: A real-world large-scale dataset for vehicle-to-vehicle cooperative perception

    Runsheng Xu, Xin Xia, Jinlong Li, Hanzhao Li, Shuo Zhang, Zhengzhong Tu, Zonglin Meng, Hao Xiang, Xiaoyu Dong, Rui Song, et al. V2v4real: A real-world large-scale dataset for vehicle-to-vehicle cooperative perception. In CVPR, 2023. 1, 3, 6, 7, 8, 14

    2023

  74. [82]

    Second: Sparsely em- bedded convolutional detection

    Yan Yan, Yuxing Mao, and Bo Li. Second: Sparsely em- bedded convolutional detection. Sensors, 18(10):3337, 2018. 4

    2018

  75. [83]

    Proposal- contrast: Unsupervised pre-training for lidar-based 3d object detection

    Junbo Yin, Dingfu Zhou, Liangjun Zhang, Jin Fang, Cheng- Zhong Xu, Jianbing Shen, and Wenguan Wang. Proposal- contrast: Unsupervised pre-training for lidar-based 3d object detection. In ECCV, 2022. 8

    2022

  76. [84]

    Learning 3d perception from others’ predictions

    Jinsu Yoo, Zhenyang Feng, Tai-Yu Pan, Yihong Sun, Cheng Perng Phoo, Xiangyu Chen, Mark Campbell, Kilian Q Weinberger, Bharath Hariharan, and Wei-Lun Chao. Learning 3d perception from others’ predictions. In ICLR, 2025. 1

    2025

  77. [85]

    Hindsight is 20/20: Leveraging past traversals to aid 3d perception

    Yurong You, Katie Z Luo, Xiangyu Chen, Junan Chen, Wei- Lun Chao, Wen Sun, Bharath Hariharan, Mark Campbell, and Kilian Q Weinberger. Hindsight is 20/20: Leveraging past traversals to aid 3d perception. In ICLR, 2022. 8

    2022

  78. [86]

    Dair-v2x: A large-scale dataset for vehicle- infrastructure cooperative 3d object detection

    Haibao Yu, Yizhen Luo, Mao Shu, Yiyi Huo, Zebang Yang, Yifeng Shi, Zhenglong Guo, Hanyu Li, Xing Hu, Jirui Yuan, and Zaiqing Nie. Dair-v2x: A large-scale dataset for vehicle- infrastructure cooperative 3d object detection. InCVPR, 2022. 3

    2022

  79. [87]

    Efficient convolutions for real-time semantic segmentation of 3d point clouds

    Chris Zhang, Wenjie Luo, and Raquel Urtasun. Efficient convolutions for real-time semantic segmentation of 3d point clouds. In 3DV, 2018. 4

    2018

  80. [88]

    Nerf-lidar: Generating realistic lidar point clouds with neural radiance fields

    Junge Zhang, Feihu Zhang, Shaochen Kuang, and Li Zhang. Nerf-lidar: Generating realistic lidar point clouds with neural radiance fields. In AAAI, 2024. 13

    2024

  81. [89]

    Adding conditional control to text-to-image diffusion models

    Lvmin Zhang, Anyi Rao, and Maneesh Agrawala. Adding conditional control to text-to-image diffusion models. In CVPR, 2023. 12

    2023

  82. [90]

    Lidar4d: Dynamic neural fields for novel space-time view lidar synthesis

    Zehan Zheng, Fan Lu, Weiyi Xue, Guang Chen, and Changjun Jiang. Lidar4d: Dynamic neural fields for novel space-time view lidar synthesis. In CVPR, 2024. 13

    2024

  83. [91]

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

    Yin Zhou and Oncel Tuzel. V oxelnet: End-to-end learning for point cloud based 3d object detection. In CVPR, 2018. 4

    2018

  84. [92]

    Rethinking pre- training and self-training

    Barret Zoph, Golnaz Ghiasi, Tsung-Yi Lin, Yin Cui, Hanx- iao Liu, Ekin Dogus Cubuk, and Quoc Le. Rethinking pre- training and self-training. In NeurIPS, 2020. 6

    2020

  85. [93]

    Un- supervised domain adaptation for semantic segmentation via class-balanced self-training

    Yang Zou, Zhiding Yu, BVK Kumar, and Jinsong Wang. Un- supervised domain adaptation for semantic segmentation via class-balanced self-training. In ECCV, 2018. 6

    2018

  86. [94]

    Learning to generate realistic lidar point clouds

    Vlas Zyrianov, Xiyue Zhu, and Shenlong Wang. Learning to generate realistic lidar point clouds. In ECCV, 2022. 2, 3, 12

    2022

  87. [95]

    Lidardm: Generative lidar simulation in a generated world

    Vlas Zyrianov, Henry Che, Zhijian Liu, and Shenlong Wang. Lidardm: Generative lidar simulation in a generated world. arXiv preprint arXiv:2404.02903, 2024. 2, 3, 12 11 In this appendix, we provide more details and experiment results in addition to the main paper: • App. A: pro...

    2024

  88. [96]

    This stage grounds the generation process, ensuring that the outputs align with given semantic cues

    during the second stage of our framework. This stage grounds the generation process, ensuring that the outputs align with given semantic cues. Domain Adaptation. Unsupervised domain adaptation (UDA) has been extensively studied. A common approach for domain adaptation is to le...

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.