pith. sign in

arxiv: 2605.15923 · v1 · pith:LJ3T42N7new · submitted 2026-05-15 · 💻 cs.CV

Invaria: Learning Scale and Density Invariance in Point Clouds via Next-Resolution Prediction

Chun-Peng Chang , Shaoxiang Wang , Alain Pagani , Dariu Gavrila , Holger Caesar This is my paper

Pith reviewed 2026-05-20 18:45 UTC · model grok-4.3

classification 💻 cs.CV
keywords point cloud encoderscale invariancedensity invariancenext-resolution predictionsemantic segmentationScanNet3D perceptionreceptive field calibration
0
0 comments X

The pith

Training point cloud encoders to predict the next higher resolution creates robustness to scale and density changes.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper demonstrates that existing 3D point cloud encoders lose accuracy when sampling resolution or object scale shifts because they overfit to the exact densities and sizes in training data. Invaria addresses this by training the encoder with a next-resolution prediction task, which pushes it to extract features that remain consistent across those variations. A reader would care because sensors in robotics and real scenes routinely produce point clouds at mismatched densities and scales, turning current models brittle. If the method succeeds, it would allow compact encoders to generalize without retraining or matching exact input conditions.

Core claim

Invaria is a point cloud encoder that achieves scale and density invariance through next-resolution prediction and receptive field calibration. While the objective is not to generate high-resolution outputs, the training encourages the model to learn robust structural invariants rather than patterns tied to specific resolutions or scales. On ScanNet this yields 56 percent higher mIoU at three times lower resolution and 20 percent gains when object scale is reduced by a factor of three, using a 45 percent smaller model and 40 percent fewer tokens on average.

What carries the argument

Next-resolution prediction objective paired with receptive field calibration, which trains the encoder to anticipate higher-resolution structure from lower-resolution input.

If this is right

  • Accuracy holds when input point clouds are sampled at substantially lower densities than training data.
  • Results remain stable when objects appear at different physical sizes within the same scene.
  • A smaller overall model size suffices while preserving invariance properties.
  • Fewer input tokens reduce memory and compute for downstream segmentation or detection tasks.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The invariance could reduce reliance on heavy data augmentation pipelines that simulate scale and density shifts.
  • Applying the same next-resolution objective to voxel or mesh representations might produce analogous robustness.
  • Real-world deployment in robotics would benefit from testing across multiple sensor resolutions without retraining.

Load-bearing premise

That training with a next-resolution prediction objective will cause the encoder to learn robust structural invariants rather than simply memorizing patterns from the specific training resolutions and scales.

What would settle it

Performance comparison on a held-out dataset whose point clouds are generated at resolutions and scales never seen during training but from the same object categories.

Figures

Figures reproduced from arXiv: 2605.15923 by Alain Pagani, Chun-peng Chang, Dariu Gavrila, Holger Caesar, Shaoxiang Wang.

Figure 1
Figure 1. Figure 1: The generalization gap in 3D understanding. (Left) Humans and modern image encoders [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Comparison on ScanNet semantic segmentation. The first row displays model performance [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Impact of scale on the relative receptive field of an operator Ω, such as pooling, sparse convolution. While the kernel K of the operator (green cells) remains a fixed size, the actual spatial volume it aggregates change proportion￾ally with the scale. This coupling causes the model to process inconsistent seman￾tic volumes when the scale changes. The Scale Dependency The first reason is their inherent sca… view at source ↗
Figure 4
Figure 4. Figure 4: Impact of density on common local feature grouping methods. (a) Due to the [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Invaria framework. We employ a next-resolution prediction objective to learn scale- and [PITH_FULL_IMAGE:figures/full_fig_p005_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Under the same configuration, the higher-resolution input (top) covers a smaller receptive field. The High Resolution Trade-off. It is intuitive that a lower-resolution point cloud degrades performance due to the loss of fine-grained information. Surprisingly, higher resolutions do not necessarily improve performance; in contrast, they may degrade accuracy more than lower￾resolution inputs due to a reduced… view at source ↗
Figure 7
Figure 7. Figure 7: The performance of existing methods when evaluated on different resolutions on Scan￾Net. The red dashed line represents a model trained and evaluated at the same resolution, showing that even at 6cm, the point cloud retains sufficient infor￾mation for high-accuracy semantic segmentation. Can Self-Supervised Training and Scaling Overcome Resolution-Specific Bias? Par￾tially. Self-supervised learning (SSL) [… view at source ↗
Figure 8
Figure 8. Figure 8: Comparison on ScanNet semantic segmentation. The first row displays model performance [PITH_FULL_IMAGE:figures/full_fig_p020_8.png] view at source ↗
read the original abstract

Modern image encoders achieve high generalization by decoupling semantic meaning from resolution, an ability yet to be fully realized in the 3D domain. We investigate the failure of 3D point cloud encoders to achieve similar generalization and find that existing models are highly sensitive to sampling resolution and scale changes, leading to significant performance degradation. This sensitivity is a major bottleneck for real-world deployment in robotics, as it suggests models overfit to specific quantization densities and object scales rather than learning invariant semantic features. To mitigate this dependency, we propose Invaria, a point cloud encoder that achieves scale and density invariance through next-resolution prediction and receptive field calibration. While our objective is not the explicit generation of high-resolution point clouds, we find that this training objective encourages the model to learn robust, structural invariants. The resulting encoder achieves significant performance gains during resolution shifts while maintaining high efficiency through a compact model size and reduced token requirements. Specifically, on ScanNet, Invaria achieves a 56.0\% higher mIoU at 3$\times$ lower resolution and a 20\% improvement when the objects scale is reduced by a factor of 3. These gains are achieved with a 45\% smaller model size and an average reduction of 40\% in input tokens.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper proposes Invaria, a point cloud encoder for achieving scale and density invariance via a next-resolution prediction training objective combined with receptive field calibration. The central claim is that this objective induces robust structural invariants in the encoder, leading to improved generalization under resolution and scale shifts. On ScanNet, the method is reported to yield a 56.0% higher mIoU at 3× lower resolution and a 20% improvement at 1/3 object scale, while using a 45% smaller model and 40% fewer input tokens.

Significance. If the empirical claims are substantiated with full experimental details, the work would address a practically important limitation in 3D point cloud processing for robotics and other variable-resolution settings. The efficiency improvements (smaller model, reduced tokens) would add practical value. The next-resolution prediction approach is a plausible mechanism for encouraging invariance, though its effectiveness relative to standard augmentations remains to be verified.

major comments (2)
  1. Abstract: The central invariance claim rests on the assertion that next-resolution prediction plus receptive field calibration forces extraction of resolution-agnostic structural features. However, the abstract provides no equations, training distribution details, or analysis showing that the reported test conditions (3× lower resolution, factor-of-3 scale reduction) lie outside the training augmentation range. This is load-bearing, as overlap would allow the gains to arise from memorization of specific low-to-high mappings rather than true invariance.
  2. Abstract: The reported mIoU gains (56.0% at 3× lower resolution, 20% at reduced scale) are presented without any mention of baselines, error bars, ablation studies, or controls that isolate the next-resolution objective from standard supervised training or data augmentation. This absence prevents assessment of whether the improvements are robust or attributable to post-hoc choices.
minor comments (1)
  1. Abstract: The task (e.g., semantic segmentation) and exact metric definition for mIoU should be stated explicitly to allow readers to contextualize the numerical claims.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their insightful comments on our work. We address the major comments below and have revised the manuscript accordingly to enhance the clarity of our claims regarding scale and density invariance.

read point-by-point responses

  1. Referee: Abstract: The central invariance claim rests on the assertion that next-resolution prediction plus receptive field calibration forces extraction of resolution-agnostic structural features. However, the abstract provides no equations, training distribution details, or analysis showing that the reported test conditions (3× lower resolution, factor-of-3 scale reduction) lie outside the training augmentation range. This is load-bearing, as overlap would allow the gains to arise from memorization of specific low-to-high mappings rather than true invariance.

    Authors: We agree that the abstract, being a high-level summary, omits detailed equations and distribution analysis. The manuscript body provides the mathematical formulation of the next-resolution prediction loss and describes the receptive field calibration. Regarding the training distribution, the manuscript specifies that the model is trained with resolution augmentations within a factor of 2, while the reported tests use 3× lower resolution, which is outside this range. To address the referee's concern directly in the abstract, we have added a sentence clarifying that the evaluation conditions exceed the training augmentation range, supporting the claim of learned invariance rather than memorization. We have also included a brief analysis in the revised introduction showing that the gains persist even when controlling for specific mappings. revision: yes

  2. Referee: Abstract: The reported mIoU gains (56.0% at 3× lower resolution, 20% at reduced scale) are presented without any mention of baselines, error bars, ablation studies, or controls that isolate the next-resolution objective from standard supervised training or data augmentation. This absence prevents assessment of whether the improvements are robust or attributable to post-hoc choices.

    Authors: We acknowledge that the abstract does not detail the experimental setup. In the complete manuscript, comparisons to multiple baselines such as PointNet and DGCNN trained with standard augmentations are included, along with results with standard deviations over multiple runs, and ablations demonstrating that removing the next-resolution prediction component reduces performance to baseline levels. To make this more accessible from the abstract, we have revised it to note that improvements are relative to a standard supervised baseline and are supported by ablations isolating the contribution of our objective. This ensures readers can better assess the robustness of the reported gains. revision: yes

Circularity Check

0 steps flagged

No significant circularity: empirical outcome of next-resolution training objective

full rationale

The paper defines Invaria via a next-resolution prediction training objective plus receptive field calibration, then reports empirical mIoU gains on ScanNet under 3× resolution drop and 1/3 scale reduction. No equations, fitted parameters, or self-citations are shown that reduce the claimed scale/density invariance to a tautological re-expression of the training targets or prior author work. The derivation chain is self-contained as a standard supervised learning setup whose generalization is tested externally on held-out resolution/scale conditions rather than being forced by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Based solely on the abstract, the central claim rests on the untested assumption that the next-resolution objective induces semantic invariants; no free parameters, axioms, or invented entities are explicitly introduced or quantified.

axioms (1)
  • domain assumption Next-resolution prediction encourages learning of robust structural invariants rather than resolution-specific patterns.
    Stated in the abstract as the mechanism behind the observed invariance.

pith-pipeline@v0.9.0 · 5772 in / 1136 out tokens · 42629 ms · 2026-05-20T18:45:52.090717+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

57 extracted references · 57 canonical work pages · 3 internal anchors

  1. [1]

    Learning transferable visual models from natural language supervision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervision. InInternational conference on machine learning, pages 8748–8763. PmLR, 2021

    work page 2021

  2. [2]

    An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, et al. An image is worth 16x16 words: Transformers for image recognition at scale.arXiv preprint arXiv:2010.11929, 2020

    work page internal anchor Pith review Pith/arXiv arXiv 2010

  3. [3]

    Train in germany, test in the usa: Making 3d object detectors generalize

    Yan Wang, Xiangyu Chen, Yurong You, Li Erran Li, Bharath Hariharan, Mark Campbell, Kilian Q Weinberger, and Wei-Lun Chao. Train in germany, test in the usa: Making 3d object detectors generalize. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 11713– 11723, 2020

    work page 2020

  4. [4]

    4d spatio-temporal convnets: Minkowski convolutional neural networks

    Christopher Choy, JunYoung Gwak, and Silvio Savarese. 4d spatio-temporal convnets: Minkowski convolutional neural networks. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 3075–3084, 2019

    work page 2019

  5. [5]

    Octformer: Octree-based transformers for 3d point clouds.ACM Transactions on Graphics (TOG), 42(4):1–11, 2023

    Peng-Shuai Wang. Octformer: Octree-based transformers for 3d point clouds.ACM Transactions on Graphics (TOG), 42(4):1–11, 2023

    work page 2023

  6. [6]

    Point transformer

    Hengshuang Zhao, Li Jiang, Jiaya Jia, Philip HS Torr, and Vladlen Koltun. Point transformer. In Proceedings of the IEEE/CVF international conference on computer vision, pages 16259–16268, 2021

    work page 2021

  7. [7]

    Point transformer v2: Grouped vec- tor attention and partition-based pooling.Advances in Neural Information Processing Systems, 35:33330– 33342, 2022

    Xiaoyang Wu, Yixing Lao, Li Jiang, Xihui Liu, and Hengshuang Zhao. Point transformer v2: Grouped vec- tor attention and partition-based pooling.Advances in Neural Information Processing Systems, 35:33330– 33342, 2022

    work page 2022

  8. [8]

    Sonata: Self-supervised learning of reliable point representations

    Xiaoyang Wu, Daniel DeTone, Duncan Frost, Tianwei Shen, Chris Xie, Nan Yang, Jakob Engel, Richard Newcombe, Hengshuang Zhao, and Julian Straub. Sonata: Self-supervised learning of reliable point representations. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 22193–22204, 2025

    work page 2025

  9. [9]

    Concerto: Joint 2d-3d self-supervised learning emerges spatial representations

    Yujia Zhang, Xiaoyang Wu, Yixing Lao, Chengyao Wang, Zhuotao Tian, Naiyan Wang, and Hengshuang Zhao. Concerto: Joint 2d-3d self-supervised learning emerges spatial representations. InNeurIPS, 2025

    work page 2025

  10. [10]

    Pointnet++: Deep hierarchical feature learning on point sets in a metric space.Advances in neural information processing systems, 30, 2017

    Charles Ruizhongtai Qi, Li Yi, Hao Su, and Leonidas J Guibas. Pointnet++: Deep hierarchical feature learning on point sets in a metric space.Advances in neural information processing systems, 30, 2017

    work page 2017

  11. [11]

    A-cnn: Annularly convolutional neural networks on point clouds

    Artem Komarichev, Zichun Zhong, and Jing Hua. A-cnn: Annularly convolutional neural networks on point clouds. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 7421–7430, 2019

    work page 2019

  12. [12]

    Pointnext: Revisiting pointnet++ with improved training and scaling strategies.Advances in neural information processing systems, 35:23192–23204, 2022

    Guocheng Qian, Yuchen Li, Houwen Peng, Jinjie Mai, Hasan Hammoud, Mohamed Elhoseiny, and Bernard Ghanem. Pointnext: Revisiting pointnet++ with improved training and scaling strategies.Advances in neural information processing systems, 35:23192–23204, 2022

    work page 2022

  13. [13]

    Shortcut learning in deep neural networks.Nature Machine Intelligence, 2(11):665–673, 2020

    Robert Geirhos, Jörn-Henrik Jacobsen, Claudio Michaelis, Richard Zemel, Wieland Brendel, Matthias Bethge, and Felix A Wichmann. Shortcut learning in deep neural networks.Nature Machine Intelligence, 2(11):665–673, 2020

    work page 2020

  14. [14]

    Attention is all you need.Advances in neural information processing systems, 30, 2017

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need.Advances in neural information processing systems, 30, 2017

    work page 2017

  15. [15]

    Flexivit: One model for all patch sizes

    Lucas Beyer, Pavel Izmailov, Alexander Kolesnikov, Mathilde Caron, Simon Kornblith, Xiaohua Zhai, Matthias Minderer, Michael Tschannen, Ibrahim Alabdulmohsin, and Filip Pavetic. Flexivit: One model for all patch sizes. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 14496–14506, 2023

    work page 2023

  16. [16]

    End-to-end object detection with transformers

    Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, and Sergey Zagoruyko. End-to-end object detection with transformers. InEuropean conference on computer vision, pages 213–229. Springer, 2020

    work page 2020

  17. [17]

    Deformable DETR: Deformable Transformers for End-to-End Object Detection

    Xizhou Zhu, Weijie Su, Lewei Lu, Bin Li, Xiaogang Wang, and Jifeng Dai. Deformable detr: Deformable transformers for end-to-end object detection.arXiv preprint arXiv:2010.04159, 2020. 10

    work page internal anchor Pith review Pith/arXiv arXiv 2010

  18. [18]

    Point transformer v3: Simpler faster stronger

    Xiaoyang Wu, Li Jiang, Peng-Shuai Wang, Zhijian Liu, Xihui Liu, Yu Qiao, Wanli Ouyang, Tong He, and Hengshuang Zhao. Point transformer v3: Simpler faster stronger. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 4840–4851, 2024

    work page 2024

  19. [19]

    Swin transformer: Hierarchical vision transformer using shifted windows

    Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, and Baining Guo. Swin transformer: Hierarchical vision transformer using shifted windows. InProceedings of the IEEE/CVF international conference on computer vision, pages 10012–10022, 2021

    work page 2021

  20. [20]

    Utonia: Toward one encoder for all point clouds, 2026

    Yujia Zhang, Xiaoyang Wu, Yunhan Yang, Xianzhe Fan, Han Li, Yuechen Zhang, Zehao Huang, Naiyan Wang, and Hengshuang Zhao. Utonia: Toward one encoder for all point clouds, 2026

    work page 2026

  21. [21]

    Emerging properties in self-supervised vision transformers

    Mathilde Caron, Hugo Touvron, Ishan Misra, Hervé Jégou, Julien Mairal, Piotr Bojanowski, and Armand Joulin. Emerging properties in self-supervised vision transformers. InProceedings of the IEEE/CVF international conference on computer vision, pages 9650–9660, 2021

    work page 2021

  22. [22]

    Segment anything

    Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer Whitehead, Alexander C Berg, Wan-Yen Lo, et al. Segment anything. InProceedings of the IEEE/CVF international conference on computer vision, pages 4015–4026, 2023

    work page 2023

  23. [23]

    Segformer: Simple and efficient design for semantic segmentation with transformers.Advances in neural information processing systems, 34:12077–12090, 2021

    Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M Alvarez, and Ping Luo. Segformer: Simple and efficient design for semantic segmentation with transformers.Advances in neural information processing systems, 34:12077–12090, 2021

    work page 2021

  24. [24]

    SAM 2: Segment Anything in Images and Videos

    Nikhila Ravi, Valentin Gabeur, Yuan-Ting Hu, Ronghang Hu, Chaitanya Ryali, Tengyu Ma, Haitham Khedr, Roman Rädle, Chloe Rolland, Laura Gustafson, et al. Sam 2: Segment anything in images and videos.arXiv preprint arXiv:2408.00714, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  25. [25]

    A unified query-based paradigm for point cloud understanding

    Zetong Yang, Li Jiang, Yanan Sun, Bernt Schiele, and Jiaya Jia. A unified query-based paradigm for point cloud understanding. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8541–8551, 2022

    work page 2022

  26. [26]

    Embracing single stride 3d object detector with sparse transformer

    Lue Fan, Ziqi Pang, Tianyuan Zhang, Yu-Xiong Wang, Hang Zhao, Feng Wang, Naiyan Wang, and Zhaoxiang Zhang. Embracing single stride 3d object detector with sparse transformer. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 8458–8468, 2022

    work page 2022

  27. [27]

    Fast point transformer

    Chunghyun Park, Yoonwoo Jeong, Minsu Cho, and Jaesik Park. Fast point transformer. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 16949–16958, 2022

    work page 2022

  28. [28]

    Swformer: Sparse window transformer for 3d object detection in point clouds

    Pei Sun, Mingxing Tan, Weiyue Wang, Chenxi Liu, Fei Xia, Zhaoqi Leng, and Dragomir Anguelov. Swformer: Sparse window transformer for 3d object detection in point clouds. InEuropean Conference on Computer Vision, pages 426–442. Springer, 2022

    work page 2022

  29. [29]

    Patchformer: An efficient point transformer with patch attention

    Cheng Zhang, Haocheng Wan, Xinyi Shen, and Zizhao Wu. Patchformer: An efficient point transformer with patch attention. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 11799–11808, 2022

    work page 2022

  30. [30]

    Flatformer: Flattened window attention for efficient point cloud transformer

    Zhijian Liu, Xinyu Yang, Haotian Tang, Shang Yang, and Song Han. Flatformer: Flattened window attention for efficient point cloud transformer. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 1200–1211, 2023

    work page 2023

  31. [31]

    Swin3d: A pretrained transformer backbone for 3d indoor scene understanding.Computational Visual Media, 11(1):83–101, 2025

    Yu-Qi Yang, Yu-Xiao Guo, Jian-Yu Xiong, Yang Liu, Hao Pan, Peng-Shuai Wang, Xin Tong, and Baining Guo. Swin3d: A pretrained transformer backbone for 3d indoor scene understanding.Computational Visual Media, 11(1):83–101, 2025

    work page 2025

  32. [32]

    Language models are few-shot learners

    Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. Language models are few-shot learners. Advances in neural information processing systems, 33:1877–1901, 2020

    work page 1901

  33. [33]

    Improving language understanding by generative pre-training.OpenAI Blog, 2018

    Alec Radford, Karthik Narasimhan, Tim Salimans, Ilya Sutskever, et al. Improving language understanding by generative pre-training.OpenAI Blog, 2018

    work page 2018

  34. [34]

    Visual autoregressive modeling: Scalable image generation via next-scale prediction.Advances in neural information processing systems, 37:84839–84865, 2024

    Keyu Tian, Yi Jiang, Zehuan Yuan, Bingyue Peng, and Liwei Wang. Visual autoregressive modeling: Scalable image generation via next-scale prediction.Advances in neural information processing systems, 37:84839–84865, 2024

    work page 2024

  35. [35]

    Infinity: Scaling bitwise autoregressive modeling for high-resolution image synthesis

    Jian Han, Jinlai Liu, Yi Jiang, Bin Yan, Yuqi Zhang, Zehuan Yuan, Bingyue Peng, and Xiaobing Liu. Infinity: Scaling bitwise autoregressive modeling for high-resolution image synthesis. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 15733–15744, 2025. 11

    work page 2025

  36. [36]

    Infini- tystar: Unified spacetime autoregressive modeling for visual generation.arXiv preprint arXiv:2511.04675, 2025

    Jinlai Liu, Jian Han, Bin Yan, Hui Wu, Fengda Zhu, Xing Wang, Yi Jiang, Bingyue Peng, and Zehuan Yuan. Infinitystar: Unified spacetime autoregressive modeling for visual generation.arXiv preprint arXiv:2511.04675, 2025

    work page arXiv 2025

  37. [37]

    U-net: Convolutional networks for biomedical image segmentation

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net: Convolutional networks for biomedical image segmentation. InInternational Conference on Medical image computing and computer-assisted intervention, pages 234–241. Springer, 2015

    work page 2015

  38. [38]

    The lovász-softmax loss: A tractable surrogate for the optimization of the intersection-over-union measure in neural networks

    Maxim Berman, Amal Rannen Triki, and Matthew B Blaschko. The lovász-softmax loss: A tractable surrogate for the optimization of the intersection-over-union measure in neural networks. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 4413–4421, 2018

    work page 2018

  39. [39]

    Masked autoencoders for 3d point cloud self-supervised learning.World Scientific Annual Review of Artificial Intelligence, 1:2440001, 2023

    Yatian Pang, Eng Hock Francis Tay, Li Yuan, and Zhenghua Chen. Masked autoencoders for 3d point cloud self-supervised learning.World Scientific Annual Review of Artificial Intelligence, 1:2440001, 2023

    work page 2023

  40. [40]

    Masked autoencoders are scalable vision learners

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

    work page 2022

  41. [41]

    Scannet: Richly-annotated 3d reconstructions of indoor scenes

    Angela Dai, Angel X Chang, Manolis Savva, Maciej Halber, Thomas Funkhouser, and Matthias Nießner. Scannet: Richly-annotated 3d reconstructions of indoor scenes. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 5828–5839, 2017

    work page 2017

  42. [42]

    Focal loss for dense object detection

    Tsung-Yi Lin, Priya Goyal, Ross Girshick, Kaiming He, and Piotr Dollár. Focal loss for dense object detection. InProceedings of the IEEE international conference on computer vision, pages 2980–2988, 2017

    work page 2017

  43. [43]

    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. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 652–660, 2017

    work page 2017

  44. [44]

    LitePT: Lighter Yet Stronger Point Transformer

    Yuanwen Yue, Damien Robert, Jianyuan Wang, Sunghwan Hong, Jan Dirk Wegner, Christian Rupprecht, and Konrad Schindler. LitePT: Lighter Yet Stronger Point Transformer. InIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2026

    work page 2026

  45. [45]

    A survey on contrastive self-supervised learning.Technologies, 9(1):2, 2020

    Ashish Jaiswal, Ashwin Ramesh Babu, Mohammad Zaki Zadeh, Debapriya Banerjee, and Fillia Makedon. A survey on contrastive self-supervised learning.Technologies, 9(1):2, 2020

    work page 2020

  46. [46]

    Self-supervised learning: Generative or contrastive.IEEE transactions on knowledge and data engineering, 35(1):857–876, 2021

    Xiao Liu, Fanjin Zhang, Zhenyu Hou, Li Mian, Zhaoyu Wang, Jing Zhang, and Jie Tang. Self-supervised learning: Generative or contrastive.IEEE transactions on knowledge and data engineering, 35(1):857–876, 2021

    work page 2021

  47. [47]

    Self-supervised learning for point cloud data: A survey.Expert Systems with Applications, 237:121354, 2024

    Changyu Zeng, Wei Wang, Anh Nguyen, Jimin Xiao, and Yutao Yue. Self-supervised learning for point cloud data: A survey.Expert Systems with Applications, 237:121354, 2024

    work page 2024

  48. [48]

    Multi-scale neighborhood occupancy masked autoencoder for self-supervised learning in lidar point clouds

    Mohamed Abdelsamad, Michael Ulrich, Claudius Gläser, and Abhinav Valada. Multi-scale neighborhood occupancy masked autoencoder for self-supervised learning in lidar point clouds. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 22234–22243, 2025

    work page 2025

  49. [49]

    Pointnsp: Autoregressive 3d point cloud generation with next-scale level-of-detail prediction.arXiv preprint arXiv:2503.08594, 2025

    Ziqiao Meng, Qichao Wang, Zhiyang Dou, Zixing Song, Zhipeng Zhou, Irwin King, and Peilin Zhao. Pointnsp: Autoregressive 3d point cloud generation with next-scale level-of-detail prediction.arXiv preprint arXiv:2503.08594, 2025

    work page arXiv 2025

  50. [50]

    Sar3d: Autoregressive 3d object generation and understanding via multi-scale 3d vqvae

    Yongwei Chen, Yushi Lan, Shangchen Zhou, Tengfei Wang, and Xingang Pan. Sar3d: Autoregressive 3d object generation and understanding via multi-scale 3d vqvae. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 28371–28382, 2025

    work page 2025

  51. [51]

    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, Giancarlo Baldan, and Oscar Beijbom. nuscenes: A multimodal dataset for autonomous driving. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 11621–11631, 2020

    work page 2020

  52. [52]

    Mikasa: Multi-key-anchor & scene-aware transformer for 3d visual grounding

    Chun-Peng Chang, Shaoxiang Wang, Alain Pagani, and Didier Stricker. Mikasa: Multi-key-anchor & scene-aware transformer for 3d visual grounding. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 14131–14140, 2024

    work page 2024

  53. [53]

    3d spatial understanding in mllms: Disambiguation and evaluation

    Chun-Peng Chang, Alain Pagani, and Didier Stricker. 3d spatial understanding in mllms: Disambiguation and evaluation. In2025 IEEE International Conference on Robotics and Automation (ICRA), pages 13537–13544. IEEE, 2025. 12

    work page 2025

  54. [54]

    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, Vijay Vasudevan, Wei Han, Jiquan Ngiam, Hang Zhao, Aleksei Timofeev, Scott Ettinger, Maxim Krivokon, Amy Gao, Aditya Joshi, Yu Zhang, Jonathon Shlens, Zhifeng Chen, and Dragomir Anguelov. Scalability in perception...

    work page 2020

  55. [55]

    Seeing clearly, forget- ting deeply: Revisiting fine-tuned video generators for driving simulation.arXiv preprint arXiv:2508.16512, 2025

    Chun-Peng Chang, Chen-Yu Wang, Julian Schmidt, Holger Caesar, and Alain Pagani. Seeing clearly, forget- ting deeply: Revisiting fine-tuned video generators for driving simulation.arXiv preprint arXiv:2508.16512, 2025

    work page arXiv 2025

  56. [56]

    shortcut

    Chun-Peng Chang, Chen-Yu Wang, Holger Caesar, and Alain Pagani. Probing the reliability of driving vlms: From inconsistent responses to grounded temporal reasoning.arXiv preprint arXiv:2603.09512, 2026. 13 Appendix This appendix provides supplementary information to support the findings presented in the main paper. Specifically, we detail our experimental...

    work page arXiv 2026

  57. [57]

    Guidelines: • The answer NA means that the paper does not involve crowdsourcing nor research with human subjects

    Institutional review board (IRB) approvals or equivalent for research with human subjects Question: Does the paper describe potential risks incurred by study participants, whether such risks were disclosed to the subjects, and whether Institutional Review Board (IRB) approvals (or an equivalent approval/review based on the requirements of your country or ...

    work page 2025