Pith. sign in

REVIEW 3 major objections 4 minor 64 references

ESMStereo: Enhanced ShuffleMixer Disparity Upsampling for Real-Time and Accurate Stereo Matching

T0 review · 3 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read ESMStereo claims that a compact cost volume plus a fast ShuffleMixer-based upsampler recovers lost detail, reaching 0.53 px EPE on SceneFlow in 26 ms and generalizing to real-world benchmarks.

desk verdict A useful real-time stereo pipeline with strong benchmark numbers, but the paper never isolates the ShuffleMixer mechanism it names; the ablation gap is real and needs fixing before publication. read the letter →

arxiv 2506.21091 v2 pith:N34H4VWR submitted 2025-06-26 cs.CV

Mahmoud Tahmasebi , Saif Huq , Kevin Meehan , Marion McAfee This is my paper
classification cs.CV
keywords stereomatchingreal-timedisparityupsamplingcostvolumeEnhancedShuffleMixerfeature-guidedrefinementedgedeployment
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

ESMStereo argues that the accuracy penalty of small-scale cost volumes in stereo matching can be mostly undone by a fast, feature-guided upsampler placed after disparity regression. The proposed Enhanced ShuffleMixer (ESM) takes the low-resolution disparity estimate, fuses it with left-image features, mixes the fused features with shuffle-and-split layers, and refines the result with a compact hourglass network. On the synthetic SceneFlow benchmark, the largest configuration reaches 0.53 pixels average endpoint error in 26 ms, beating or matching slower real-time methods, while the smallest runs at 116 FPS on a desktop GPU and 91 FPS on an embedded Jetson AGX Orin. If correct, this shifts the design trade-off: instead of paying for a large cost volume and heavy 3D aggregation, one can spend computation after regression in 2D upsampling and still get accurate, real-time disparity maps.

What carries the argument

The central mechanism is the Enhanced ShuffleMixer (ESM) module, a 2D upsampling and refinement block. It first uses four convolutional layers to extract features from the low-resolution disparity map, concatenates them with image features from the left view, processes the fusion through two FMBlocks (the shuffle-split mixing layers from ShuffleMixer), then applies a pixel shuffle to reach higher resolution and runs a compact feature-guided hourglass network to smooth and enrich the result. The design intent is to give a large receptive field and local contextual mixing at low FLOPs, shifting computation from 3D aggregation to 2D refinement.

What would settle it

Retrain ESMStereo-L-gwc on SceneFlow with the two FMBlocks in the ESM module replaced by two 3x3 convolutional layers matched in channel count and stride, keeping everything else identical; if the endpoint error stays essentially at 0.53 px, the shuffle-split mixing is not the mechanism delivering the accuracy gain. A second check is to remove the ESM module entirely and replace it with bilinear upsampling: the gap to 0.53 px then measures how much of the claimed gain actually comes from the upsampler.

Watch

Extended reading notes

Core claim

The paper's central claim is that a small-scale cost volume plus a lightweight 3D hourglass aggregation, followed by the ESM module, is sufficient to reach state-of-the-art real-time accuracy. ESM restores information lost in the compact volume by extracting features from the initial disparity map, concatenating them with image features, mixing through two FMBlocks from ShuffleMixer, upsampling with pixel shuffle, and refining through a feature-guided hourglass. The result is that ESMStereo-L-gwc achieves 0.53 px EPE on SceneFlow in 26 ms and the best generalization among real-time methods on KITTI, Middlebury, and ETH3D when trained only on SceneFlow; the module also improves PSMNet and Fast-ACVNet-Plus when inserted. The upsampling stage, not the cost volume, carries the burden of recovering fine structure.

Load-bearing premise

The whole accuracy story depends on the ESM upsampler being the thing that recovers lost detail, but the paper never tests whether its shuffle-split mixing layers beat plain convolutional layers of similar cost.

Editorial extensions

If this is right

  • Compact cost volumes at 1/4 to 1/16 resolution suffice for near-state-of-the-art accuracy when ESM upsampling follows, cutting the memory and FLOPs of full-range cost volumes.
  • Adding ESM to existing networks is a direct accuracy boost: SceneFlow EPE drops from 1.09 to 1.02 px for PSMNet and from 0.59 to 0.51 px for Fast-ACVNet-Plus.
  • Models trained only on synthetic data generalize to real scenes without fine-tuning, with ESMStereo-L-gwc reaching 5.5% D1 on KITTI 2015, 11.1% bad 2.0 on Middlebury, and 6.1% bad 1.0 on ETH3D.
  • On edge hardware, the smallest version runs at 91 FPS on an NVIDIA Jetson AGX Orin with 1.7 million parameters, showing the speed-accuracy trade-off persists outside desktop GPUs.
  • Within the compact-volume design, group-wise correlation volumes consistently beat norm-correlation volumes at the same resolution, so cost-volume type remains a meaningful accuracy lever.

Reading between the lines

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

  • A testable extension the paper leaves implicit: reduce the cost volume to 1/32 resolution and see whether ESM continues to absorb the information loss; the design logic predicts a graceful rather than catastrophic accuracy decline.
  • Because ESM only needs a coarse disparity estimate and a guide image, the same module could be applied to optical flow refinement or monocular depth upsampling, where a low-resolution prediction is fused with the original image.
  • The paper never ablates the ShuffleMixer FMBlocks against plain convolutional mixing; running that comparison would determine whether shuffle-split mixing is essential or merely an efficient implementation.
  • The portability results suggest ESM might generalize as a drop-in refinement stage beyond the two tested networks, but that claim would need experiments with iterative or transformer-based stereo backbones to confirm.
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

3 major / 4 minor

Summary. The paper proposes ESMStereo, a family of real-time stereo matching networks built around compact correlation cost volumes, a lightweight 3D hourglass aggregation module, and a new Enhanced ShuffleMixer (ESM) upsampling unit. The ESM module extracts disparity features, fuses them with image features, mixes them through two FMBlocks taken from ShuffleMixer, upsamples with pixel shuffle, and refines with a compact feature-guided hourglass network. The authors report SceneFlow EPE of 0.53 px for the large model, 1.10 px for the small model, inference speeds up to 116 FPS on an RTX 4070 SUPER and 91 FPS on an AGX Orin, KITTI results after fine-tuning, zero-shot generalization results on KITTI, Middlebury, and ETH3D, and a portability study showing that adding the full ESM module improves PSMNet and Fast-ACVNet-Plus.

Significance. If the reported results hold, the large ESMStereo variant offers a genuinely competitive accuracy-speed trade-off for real-time stereo matching, with unusually broad evaluation across synthetic, real-world, and edge-device settings. The paper also ships source code, which supports reproducibility. However, the significance is substantially moderated by the absence of any ablation isolating the ShuffleMixer component that the paper identifies as its core mechanism, and by claims that extend the large-model result to the entire model family. The contribution is best read as a promising engineering architecture whose specific novelty attribution is not yet established by the reported experiments.

major comments (3)
  1. [Section 4.1 / Table 2 / Section 5.7] The central claimed mechanism, the two FMBlocks from ShuffleMixer, is never ablated. Table 2 varies only the cost-volume type and resolution/backbone, while Table 7 compares whole pipelines with and without the entire ESM module. Because the ESM module introduces disparity-feature convolutions, fusion convolutions, FMBlocks, pixel shuffle, and an hourglass refinement jointly, the reported gains (e.g., Fast-ACVNet-Plus going from 0.59 to 0.51 EPE) cannot be attributed to the ShuffleMixer-style mixing. A controlled ablation that removes or replaces the FMBlocks with ordinary convolutional mixing is needed to support the paper's title and the 'Enhanced ShuffleMixer' claim.
  2. [Section 3 / Section 5.4 / Table 3] The statement that ESMStereo 'significantly surpasses' RTSMNet and FBPGNet is contradicted by the reported numbers for the smaller variants. RTSMNet-c8 has EPE 0.71 on SceneFlow, while ESMStereo-S-gwc is 1.10 and ESMStereo-M-gwc is 0.77; only ESMStereo-L-gwc (0.53) beats RTSMNet. Similarly, ESMStereo-S-gwc (1.10) is worse than SADSNet-L-N7 (0.90). The claims about state-of-the-art accuracy should be restricted to the large variant, or the presentation should explain why S and M are positioned differently.
  3. [Section 5.4 / Tables 3 and 5] The runtime comparisons mix GPUs of substantially different generations without correction: the proposed models are timed on an RTX 4070 SUPER, while most competitors are timed on an RTX 3090, RTX 2080 Ti, or Titan X. Directly comparing 26 ms on RTX 4070S with, say, 27 ms on RTX 3090 is not a clean same-hardware comparison and weakens the speed-accuracy claims. The authors should either report timings on a common GPU or clearly state that the timing comparison is indicative only.
minor comments (4)
  1. [Table 6] The rows LightStereo-M and LightStereo-L are cited as [13], but these results come from the LightStereo paper [20]; the reference citation appears to be a copy-paste error.
  2. [Section 5.6] The text refers to 'KITTI 2013' when describing the generalization evaluation; the dataset used is KITTI 2015, based on the table and earlier text.
  3. [Table 1] The layer labels are not sequential: two rows are both labeled '6' and the following row is labeled '7'. Please renumber the rows consecutively.
  4. [Section 5.8] The complexity analysis reports parameters only for the proposed models and one competitor (LightStereo-H); adding parameter counts for the other compared methods would make the efficiency comparison more informative.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity found: the paper's claims rest on external benchmarks and whole-module integration experiments, not on fitted inputs or self-citation chains.

full rationale

The paper does not derive its predictions from a fitted parameter. Its load-bearing results are measured EPE and D1 values on the external SceneFlow, KITTI, Middlebury, and ETH3D benchmarks. The ESM module is introduced as a network component in Section 4.1, and its value is tested empirically in Section 5.7 by retraining PSMNet and Fast-ACVNet-Plus with and without ESM. There is no equation or construction that makes the output equal to an input by definition. The cited ShuffleMixer FMBlocks are taken from an external source [41], and the self-citation to DCVSMNet [5] appears as a baseline comparison, not as a justification that forces the result. The lack of an ablation isolating the FMBlocks from other ESM parts is a completeness limitation, but it does not constitute circularity under the standard of exhibiting a specific reduction. Therefore the circularity score is 0.

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

No new physical or conceptual entities are introduced. The ESM module is an architectural component, not a new particle, force, dimension, or conserved quantity. The report depends only on standard learning assumptions and the paper's design choices.

free parameters (4)
  • Multi-scale loss weights lambda_i = {1, 1/6, 1/10}
    Chosen in Equation 3; weights the three ESM output stages. No sensitivity analysis is provided.
  • Top-k regression values = k=1 for 1/8 and 1/16, k=2 for 1/4
    Defined in Section 4.3 following CoEx; a design choice affecting the quality of the initial disparity estimates.
  • Number of FMBlocks in ESM = 2
    Set in Section 4.1 and Figure 3; no ablation varies the number of mixing blocks.
  • Cost volume resolution per model variant = 1/16 (S), 1/8 (M), 1/4 (L)
    Section 4.2 and Table 2; the resolution trades speed against accuracy, but the paper does not compare equal-parameter variants.
assumptions (4)
  • domain assumption Benchmark ground truth from SceneFlow, KITTI, ETH3D, and Middlebury is accurate enough to support the reported EPE and D1 comparisons.
    Section 5.1 treats these datasets as valid evaluation standards without analyzing ground-truth errors or labeling noise.
  • domain assumption Published numbers for competing methods were obtained under comparable training and testing protocols.
    Tables 3, 5, and 6 compare against numbers measured on different GPUs and potentially different input resolutions, which affects runtime comparisons and, to a lesser extent, accuracy comparisons.
  • ad hoc to paper FMBlocks from ShuffleMixer improve local feature mixing in the disparity upsampling context.
    Section 4.1 adopts the block from [41] without an ablation showing it outperforms plain convolutions in this pipeline.
  • domain assumption Training on synthetic SceneFlow transfers to real-world KITTI, Middlebury, and ETH3D sufficiently for zero-shot generalization claims.
    Section 5.2 trains on SceneFlow and Section 5.6 evaluates zero-shot; no domain-shift analysis beyond aggregate metrics is given.

how reviews work

0 comments
Cite this review

Pith. "Pith review of ESMStereo: Enhanced ShuffleMixer Disparity Upsampling for Real-Time and Accurate Stereo Matching." pith.science (2026). https://pith.science/paper/N34H4VWR

@misc{pith2026250621091,
  author       = {Pith},
  title        = {Pith review of: ESMStereo: Enhanced ShuffleMixer Disparity Upsampling for Real-Time and Accurate Stereo Matching},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/N34H4VWR}},
  note         = {Machine review of arXiv:2506.21091}
}
read the original abstract

Stereo matching has become an increasingly important component of modern autonomous systems. Developing deep learning-based stereo matching models that deliver high accuracy while operating in real-time continues to be a major challenge in computer vision. In the domain of cost-volume-based stereo matching, accurate disparity estimation depends heavily on large-scale cost volumes. However, such large volumes store substantial redundant information and also require computationally intensive aggregation units for processing and regression, making real-time performance unattainable. Conversely, small-scale cost volumes followed by lightweight aggregation units provide a promising route for real-time performance, but lack sufficient information to ensure highly accurate disparity estimation. To address this challenge, we propose the Enhanced Shuffle Mixer (ESM) to mitigate information loss associated with small-scale cost volumes. ESM restores critical details by integrating primary features into the disparity upsampling unit. It quickly extracts features from the initial disparity estimation and fuses them with image features. These features are mixed by shuffling and layer splitting then refined through a compact feature-guided hourglass network to recover more detailed scene geometry. The ESM focuses on local contextual connectivity with a large receptive field and low computational cost, leading to the reconstruction of a highly accurate disparity map at real-time. The compact version of ESMStereo achieves an inference speed of 116 FPS on high-end GPUs and 91 FPS on the AGX Orin.

Figures

Figures reproduced from arXiv: 2506.21091 by the authors.

Figure 1
Figure 1. Comparison of various versions of ESMStereo with state-of-the-art real-time methods. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. In ESMStereo, the volume is constructed at 1 4 resolution and aggregated by a lightweight 3D hourglass network to generate a low resolution initial disparity map which is further up-sampled by ESM modules to estimate the full resolution disparity map. This figure illustrates the architecture for the large version of ESMStereo a low resolution or sparse cost volume to speed up the network while mitigating accuracy lo… view at source ↗
Figure 3
Figure 3. The ESM module extracts features from the low-resolution disparity map using four 2D convolutional layers, which are subsequently fused with contextual features from the left image. The fused features are processed through two FMBlocks [41], which mix the features by shuffling and splitting the layers to enhance the focus on local correlations within the features. Following this, a feature-guided hourglass refinemen… view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Generalization results on KITTI 2012 training set. The models are trained only on the SceneFlow dataset. The left panel shows the left input image and the groundtruth dispar￾ity. For each example, the first row shows the colorized disparity prediction and the second ro…
Figure 5
Figure 5. Figure 5: Generalization results on KITTI 2015 for the models only trained on SceneFlow dataset. The left panel shows the left input image and the groundtruth disparity. For each example, the first row shows the colorized disparity prediction and the second row shows the D1 erro…
Figure 6
Figure 6. Figure 6: Generalization results of ESMStereo on Middlebury 2014 dataset when only trained on the synthetic SceneFlow dataset. The first row shows the left image and the groundtruth, the second row represents the colorized disparity prediction and the third row shows the D1 erro…
Figure 8
Figure 8. Figure 8: Qualitative results on KITTI 2015 for the integrated ESM module. The left panel shows the left input image and the groundtruth disparity, the second column is the results without ESM and the third column is the results with ESM . For each example, the first row shows t…
Figure 7
Figure 7. Figure 7: Generalization results of ESMStereo on ETH3D dataset. The first row shows the left image and the groundtruth, the second row represents the colorized disparity prediction and the third row shows the D1 error map. 5.8 Computational Complexity Analysis The low computatio…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

64 extracted references · 53 canonical work pages

  1. [1]

    Igev++: Iterative multi-range geometry encoding volumes for stereo matching.arXiv preprint arXiv:2409.00638, 2024

    Gangwei Xu, Xianqi Wang, Zhaoxing Zhang, Junda Cheng, Chunyuan Liao, and Xin Yang. Igev++: Iterative multi-range geometry encoding volumes for stereo matching.arXiv preprint arXiv:2409.00638, 2024

    arXiv 2024

  2. [2]

    Mobilestereonet: Towards lightweight deep networks for stereo matching

    Faranak Shamsafar, Samuel Woerz, Rafia Rahim, and Andreas Zell. Mobilestereonet: Towards lightweight deep networks for stereo matching. InProceedings of the IEEE/CVF Winter Con- ference on Applications of Computer Vision, pages 2417–2426, 2022

    work page 2022

  3. [3]

    Correlate-and-excite: Real-time stereo matching via guided cost volume excita- tion

    Antyanta Bangunharcana, Jae Won Cho, Seokju Lee, In So Kweon, Kyung-Soo Kim, and Soohyun Kim. Correlate-and-excite: Real-time stereo matching via guided cost volume excita- tion. In2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 3542–3548. IEEE, 2021

    work page 2021

  4. [4]

    Feature-guided spatial attention upsampling for real- time stereo matching network.IEEE MultiMedia, 28(1):38–47, 2020

    Yun Xie, Shaowu Zheng, and Weihua Li. Feature-guided spatial attention upsampling for real- time stereo matching network.IEEE MultiMedia, 28(1):38–47, 2020

    work page 2020

  5. [5]

    Dcvsmnet: Double cost volume stereo matching network.Neurocomputing, page 129002, 2024

    Mahmoud Tahmasebi, Saif Huq, Kevin Meehan, and Marion McAfee. Dcvsmnet: Double cost volume stereo matching network.Neurocomputing, page 129002, 2024

    work page 2024

  6. [6]

    Autodisp- net: Improving disparity estimation with automl

    Tonmoy Saikia, Yassine Marrakchi, Arber Zela, Frank Hutter, and Thomas Brox. Autodisp- net: Improving disparity estimation with automl. InProceedings of the ieee/cvf international conference on computer vision, pages 1812–1823, 2019

    work page 2019

  7. [7]

    Accurate and Efficient Stereo Matching via Attention Concatenation Volume

    Gangwei Xu, Yun Wang, Junda Cheng, Jinhui Tang, and Xin Yang. Accurate and efficient stereo matching via attention concatenation volume.arXiv preprint arXiv:2209.12699, 2022

    work page Pith review arXiv 2022

  8. [8]

    Efficientnetv2: Smaller models and faster training

    Mingxing Tan and Quoc Le. Efficientnetv2: Smaller models and faster training. InInternational conference on machine learning, pages 10096–10106. PMLR, 2021

    2021

Show all 64 references
  1. [9]

    Mocha-stereo: Motif channel attention network for stereo matching.arXiv preprint arXiv:2404.06842, 2024

    Ziyang Chen, Wei Long, He Yao, Yongjun Zhang, Bingshu Wang, Yongbin Qin, and Jia Wu. Mocha-stereo: Motif channel attention network for stereo matching.arXiv preprint arXiv:2404.06842, 2024

    2024 arXiv

  2. [10]

    Iterative geometry encoding volume for stereo matching

    Gangwei Xu, Xianqi Wang, Xiaohuan Ding, and Xin Yang. Iterative geometry encoding volume for stereo matching. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 21919–21928, 2023

    2023

  3. [11]

    Diffuvolume: Diffu- sion model for volume based stereo matching.arXiv preprint arXiv:2308.15989, 2023

    Dian Zheng, Xiao-Ming Wu, Zuhao Liu, Jingke Meng, and Wei-shi Zheng. Diffuvolume: Diffu- sion model for volume based stereo matching.arXiv preprint arXiv:2308.15989, 2023

    2023 arXiv

  4. [12]

    Stereonet: Guided hierarchical refinement for real-time edge-aware depth prediction

    Sameh Khamis, Sean Fanello, Christoph Rhemann, Adarsh Kowdle, Julien Valentin, and Shahram Izadi. Stereonet: Guided hierarchical refinement for real-time edge-aware depth prediction. InProceedings of the European Conference on Computer Vision (ECCV), pages 573–590, 2018

    2018

  5. [13]

    Cgi-stereo: Accurate and real-time stereo matching via context and geometry interaction.arXiv preprint arXiv:2301.02789, 2023

    Gangwei Xu, Huan Zhou, and Xin Yang. Cgi-stereo: Accurate and real-time stereo matching via context and geometry interaction.arXiv preprint arXiv:2301.02789, 2023

    2023 arXiv

  6. [14]

    Bilateral grid learning for stereo matching networks

    Bin Xu, Yuhua Xu, Xiaoli Yang, Wei Jia, and Yulan Guo. Bilateral grid learning for stereo matching networks. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 12497–12506, 2021

    2021

  7. [15]

    Ebstereo: edge-based loss function for real-time stereo matching.The Visual Computer, pages 1–12, 2023

    Weijie Bi, Ming Chen, Dongliu Wu, and Shenglian Lu. Ebstereo: edge-based loss function for real-time stereo matching.The Visual Computer, pages 1–12, 2023. 18

    2023

  8. [16]

    Multi-dimensional cooperative net- work for stereo matching.IEEE Robotics and Automation Letters, 7(1):581–587, 2021

    Wei Chen, Xiaogang Jia, Mingfei Wu, and Zhengfa Liang. Multi-dimensional cooperative net- work for stereo matching.IEEE Robotics and Automation Letters, 7(1):581–587, 2021

    2021

  9. [17]

    Deep- pruner: Learning efficient stereo matching via differentiable patchmatch

    Shivam Duggal, Shenlong Wang, Wei-Chiu Ma, Rui Hu, and Raquel Urtasun. Deep- pruner: Learning efficient stereo matching via differentiable patchmatch. InProceedings of the IEEE/CVF international conference on computer vision, pages 4384–4393, 2019

    2019

  10. [18]

    Aanet: Adaptive aggregation network for efficient stereo match- ing

    Haofei Xu and Juyong Zhang. Aanet: Adaptive aggregation network for efficient stereo match- ing. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 1959–1968, 2020

    1959

  11. [19]

    Iinet: Implicit intra-inter information fusion for real-time stereo matching

    Ximeng Li, Chen Zhang, Wanjuan Su, and Wenbing Tao. Iinet: Implicit intra-inter information fusion for real-time stereo matching. InProceedings of the AAAI Conference on Artificial Intelligence, volume 38, pages 3225–3233, 2024

    2024

  12. [20]

    Lightstereo: Channel boost is all your need for efficient 2d cost aggregation.arXiv preprint arXiv:2406.19833, 2024

    Xianda Guo, Chenming Zhang, Dujun Nie, Wenzhao Zheng, Youmin Zhang, and Long Chen. Lightstereo: Channel boost is all your need for efficient 2d cost aggregation.arXiv preprint arXiv:2406.19833, 2024

    2024 arXiv

  13. [21]

    Towards accurate and real-time binocular vision: A lightweight stereo matching network for rgb stereo images.IEEE Sensors Journal, 2023

    Zhong Wu, Hong Zhu, Lili He, and Jing Shi. Towards accurate and real-time binocular vision: A lightweight stereo matching network for rgb stereo images.IEEE Sensors Journal, 2023

    2023

  14. [22]

    Detail preserving coarse-to- fine matching for stereo matching and optical flow.IEEE Transactions on Image Processing, 30:5835–5847, 2021

    Yong Deng, Jimin Xiao, Steven Zhiying Zhou, and Jiashi Feng. Detail preserving coarse-to- fine matching for stereo matching and optical flow.IEEE Transactions on Image Processing, 30:5835–5847, 2021

    2021

  15. [23]

    Edgestereo: An effective multi-task learning network for stereo matching and edge detection.International Journal of Computer Vision, 128:910–930, 2020

    Xiao Song, Xu Zhao, Liangji Fang, Hanwen Hu, and Yizhou Yu. Edgestereo: An effective multi-task learning network for stereo matching and edge detection.International Journal of Computer Vision, 128:910–930, 2020

    2020

  16. [24]

    Real-time semantic stereo matching

    Pier Luigi Dovesi, Matteo Poggi, Lorenzo Andraghetti, Miquel Mart ´ ı, Hedvig Kjellstr¨ om, Alessandro Pieropan, and Stefano Mattoccia. Real-time semantic stereo matching. In2020 IEEE international conference on robotics and automation (ICRA), pages 10780–10787. IEEE, 2020

    2020

  17. [25]

    Segstereo: Ex- ploiting semantic information for disparity estimation

    Guorun Yang, Hengshuang Zhao, Jianping Shi, Zhidong Deng, and Jiaya Jia. Segstereo: Ex- ploiting semantic information for disparity estimation. InProceedings of the European conference on computer vision (ECCV), pages 636–651, 2018

    2018

  18. [26]

    Convmlp-mixer based real-time stereo matching network towards autonomous driving.IEEE Transactions on Vehic- ular Technology, 72(2):2581–2586, 2022

    Cheng Liu, Wei Wei, Bifa Liang, Xihao Liu, Wenli Shang, and Jun Li. Convmlp-mixer based real-time stereo matching network towards autonomous driving.IEEE Transactions on Vehic- ular Technology, 72(2):2581–2586, 2022

    2022

  19. [27]

    Accurate and efficient stereo matching via attention concatenation volume.IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023

    Gangwei Xu, Yun Wang, Junda Cheng, Jinhui Tang, and Xin Yang. Accurate and efficient stereo matching via attention concatenation volume.IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023

    2023

  20. [28]

    Anytime stereo image depth estimation on mobile devices

    Yan Wang, Zihang Lai, Gao Huang, Brian H Wang, Laurens Van Der Maaten, Mark Campbell, and Kilian Q Weinberger. Anytime stereo image depth estimation on mobile devices. In2019 international conference on robotics and automation (ICRA), pages 5893–5900. IEEE, 2019

    2019

  21. [29]

    Adcpnet: Adaptive dispar- ity candidates prediction network for efficient real-time stereo matching.arXiv preprint arXiv:2011.09023, 2020

    He Dai, Xuchong Zhang, Yongli Zhao, and Hongbin Sun. Adcpnet: Adaptive dispar- ity candidates prediction network for efficient real-time stereo matching.arXiv preprint arXiv:2011.09023, 2020. 19

    2011 arXiv

  22. [30]

    Bifa Liang, Wei Wei, Jinhao Huang, Cheng Liu, Hong Yang, Ru Yang, Wenli Shang, and Jun Li. Real-time stereo image depth estimation network with group-wise l1 distance for edge devices towards autonomous driving.IEEE Transactions on Vehicular Technology, 72(11):13917–13928, 2023

    2023

  23. [31]

    Jinhao Huang, Wei Wei, Bifa Liang, Cheng Liu, Wenli Shang, and Jun Li. Real-time stereo disparity prediction based on patch-embedded extraction and depthwise hierarchical refine- ment for 3-d sensing of autonomous vehicles on energy-efficient edge computing devices.IEEE Transa...

    2023

  24. [32]

    A large dataset to train convolutional networks for disparity, optical flow, and scene flow estimation

    Nikolaus Mayer, Eddy Ilg, Philip Hausser, Philipp Fischer, Daniel Cremers, Alexey Dosovitskiy, and Thomas Brox. A large dataset to train convolutional networks for disparity, optical flow, and scene flow estimation. InProceedings of the IEEE conference on computer vision and p...

    2016

  25. [33]

    Real-time stereo matching network with high accuracy

    Hyunmin Lee and Yongho Shin. Real-time stereo matching network with high accuracy. In2019 IEEE International Conference on Image Processing (ICIP), pages 4280–4284. IEEE, 2019

    2019

  26. [34]

    Pyramid stereo matching network

    Jia-Ren Chang and Yong-Sheng Chen. Pyramid stereo matching network. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 5410–5418, 2018

    2018

  27. [35]

    Group-wise cor- relation stereo network

    Xiaoyang Guo, Kai Yang, Wukui Yang, Xiaogang Wang, and Hongsheng Li. Group-wise cor- relation stereo network. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 3273–3282, 2019

    2019

  28. [36]

    Real-time stereo matching with high accuracy via spatial attention-guided upsampling.Applied Intelligence, 53(20):24253–24274, 2023

    Zhong Wu, Hong Zhu, Lili He, Qiang Zhao, Jing Shi, and Wenhuan Wu. Real-time stereo matching with high accuracy via spatial attention-guided upsampling.Applied Intelligence, 53(20):24253–24274, 2023

    2023

  29. [37]

    Deformable convnets v2: More deformable, better results

    Xizhou Zhu, Han Hu, Stephen Lin, and Jifeng Dai. Deformable convnets v2: More deformable, better results. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 9308–9316, 2019

    2019

  30. [38]

    Carafe: Content- aware reassembly of features

    Jiaqi Wang, Kai Chen, Rui Xu, Ziwei Liu, Chen Change Loy, and Dahua Lin. Carafe: Content- aware reassembly of features. InProceedings of the IEEE/CVF international conference on computer vision, pages 3007–3016, 2019

    2019

  31. [39]

    Real-time single image and video super-resolution using an efficient sub-pixel convolutional neural network

    Wenzhe Shi, Jose Caballero, Ferenc Husz´ ar, Johannes Totz, Andrew P Aitken, Rob Bishop, Daniel Rueckert, and Zehan Wang. Real-time single image and video super-resolution using an efficient sub-pixel convolutional neural network. InProceedings of the IEEE conference on comput...

    2016

  32. [40]

    Feature back-projection guided residual refinement for real-time stereo matching network.Signal Processing: Image Commu- nication, 103:116636, 2022

    Bin Wen, Han Zhu, Chao Yang, Zhicong Li, and Renxuan Cao. Feature back-projection guided residual refinement for real-time stereo matching network.Signal Processing: Image Commu- nication, 103:116636, 2022

    2022

  33. [41]

    Shufflemixer: An efficient convnet for image super- resolution.Advances in Neural Information Processing Systems, 35:17314–17326, 2022

    Long Sun, Jinshan Pan, and Jinhui Tang. Shufflemixer: An efficient convnet for image super- resolution.Advances in Neural Information Processing Systems, 35:17314–17326, 2022

    2022

  34. [42]

    High- frequency stereo matching network

    Haoliang Zhao, Huizhou Zhou, Yongjun Zhang, Jie Chen, Yitong Yang, and Yong Zhao. High- frequency stereo matching network. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 1327–1336, 2023

    2023

  35. [43]

    Hitnet: Hierarchical iterative tile refinement network for real-time stereo matching

    Vladimir Tankovich, Christian Hane, Yinda Zhang, Adarsh Kowdle, Sean Fanello, and Sofien Bouaziz. Hitnet: Hierarchical iterative tile refinement network for real-time stereo matching. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 1...

    2021

  36. [44]

    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. In2012 IEEE conference on computer vision and pattern recognition, pages 3354–3361. IEEE, 2012

    2012

  37. [45]

    Object scene flow for autonomous vehicles

    Moritz Menze and Andreas Geiger. Object scene flow for autonomous vehicles. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 3061–3070, 2015

    2015

  38. [46]

    A multi-view stereo benchmark with high-resolution images and multi-camera videos

    Thomas Schops, Johannes L Schonberger, Silvano Galliani, Torsten Sattler, Konrad Schindler, Marc Pollefeys, and Andreas Geiger. A multi-view stereo benchmark with high-resolution images and multi-camera videos. InProceedings of the IEEE conference on computer vision and patter...

    2017

  39. [47]

    High-resolution stereo datasets with subpixel-accurate ground truth

    Daniel Scharstein, Heiko Hirschm¨ uller, York Kitajima, Greg Krathwohl, Nera Neˇ si´ c, Xi Wang, and Porter Westling. High-resolution stereo datasets with subpixel-accurate ground truth. In Pattern Recognition: 36th German Conference, GCPR 2014, M¨ unster, Germany, September 2...

    2014

  40. [48]

    Decoupled weight decay regularization.arXiv preprint arXiv:1711.05101, 2017

    I Loshchilov. Decoupled weight decay regularization.arXiv preprint arXiv:1711.05101, 2017

    2017 arXiv

  41. [49]

    Mobilenetv2: Inverted residuals and linear bottlenecks

    Mark Sandler, Andrew Howard, Menglong Zhu, Andrey Zhmoginov, and Liang-Chieh Chen. Mobilenetv2: Inverted residuals and linear bottlenecks. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 4510–4520, 2018

    2018

  42. [50]

    Fadnet++: Real-time and accurate disparity estimation with configurable networks.arXiv preprint arXiv:2110.02582, 2021

    Qiang Wang, Shaohuai Shi, Shizhen Zheng, Kaiyong Zhao, and Xiaowen Chu. Fadnet++: Real-time and accurate disparity estimation with configurable networks.arXiv preprint arXiv:2110.02582, 2021

    2021 arXiv

  43. [51]

    Cfnet: Cascade and fused cost volume for robust stereo matching

    Zhelun Shen, Yuchao Dai, and Zhibo Rao. Cfnet: Cascade and fused cost volume for robust stereo matching. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 13906–13915, 2021

    2021

  44. [52]

    Hierarchical neural architecture search for deep stereo matching.Advances in Neural Information Processing Systems, 33:22158–22169, 2020

    Xuelian Cheng, Yiran Zhong, Mehrtash Harandi, Yuchao Dai, Xiaojun Chang, Hongdong Li, Tom Drummond, and Zongyuan Ge. Hierarchical neural architecture search for deep stereo matching.Advances in Neural Information Processing Systems, 33:22158–22169, 2020

    2020

  45. [53]

    Semantic stereo matching with pyramid cost volumes

    Zhenyao Wu, Xinyi Wu, Xiaoping Zhang, Song Wang, and Lili Ju. Semantic stereo matching with pyramid cost volumes. InProceedings of the IEEE/CVF international conference on computer vision, pages 7484–7493, 2019

    2019

  46. [54]

    Learning depth with convolutional spatial prop- agation network.IEEE transactions on pattern analysis and machine intelligence, 42(10):2361– 2379, 2019

    Xinjing Cheng, Peng Wang, and Ruigang Yang. Learning depth with convolutional spatial prop- agation network.IEEE transactions on pattern analysis and machine intelligence, 42(10):2361– 2379, 2019

    2019

  47. [55]

    Ga-net: Guided aggregation net for end-to-end stereo matching

    Feihu Zhang, Victor Prisacariu, Ruigang Yang, and Philip HS Torr. Ga-net: Guided aggregation net for end-to-end stereo matching. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 185–194, 2019

    2019

  48. [56]

    Local similarity pattern and cost self-reassembling for deep stereo matching networks

    Biyang Liu, Huimin Yu, and Yangqi Long. Local similarity pattern and cost self-reassembling for deep stereo matching networks. InProceedings of the AAAI Conference on Artificial Intel- ligence, volume 36, pages 1647–1655, 2022

    2022

  49. [57]

    P3snet: Parallel pyramid pooling stereo network.IEEE Trans- actions on Intelligent Transportation Systems, 2023

    Alper Emlek and Murat Peker. P3snet: Parallel pyramid pooling stereo network.IEEE Trans- actions on Intelligent Transportation Systems, 2023

    2023

  50. [58]

    Attention concatenation volume for accurate and efficient stereo matching

    Gangwei Xu, Junda Cheng, Peng Guo, and Xin Yang. Attention concatenation volume for accurate and efficient stereo matching. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 12981–12990, 2022. 21

    2022

  51. [59]

    A decomposition model for stereo matching

    Chengtang Yao, Yunde Jia, Huijun Di, Pengxiang Li, and Yuwei Wu. A decomposition model for stereo matching. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 6091–6100, 2021

    2021

  52. [60]

    A joint 2d-3d complementary network for stereo matching.Sensors, 21(4):1430, 2021

    Xiaogang Jia, Wei Chen, Zhengfa Liang, Xin Luo, Mingfei Wu, Chen Li, Yulin He, Yusong Tan, and Libo Huang. A joint 2d-3d complementary network for stereo matching.Sensors, 21(4):1430, 2021

    2021

  53. [61]

    Domain-invariant stereo matching networks

    Feihu Zhang, Xiaojuan Qi, Ruigang Yang, Victor Prisacariu, Benjamin Wah, and Philip Torr. Domain-invariant stereo matching networks. InComputer Vision–ECCV 2020: 16th Euro- pean Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part II 16, pages 420–439. Springer, 2020

    2020

  54. [62]

    Revisiting stereo depth estimation from a sequence-to-sequence perspective with transformers

    Zhaoshuo Li, Xingtong Liu, Nathan Drenkow, Andy Ding, Francis X Creighton, Russell H Taylor, and Mathias Unberath. Revisiting stereo depth estimation from a sequence-to-sequence perspective with transformers. InProceedings of the IEEE/CVF international conference on computer v...

    2021

  55. [63]

    Revisiting domain generalized stereo matching networks from a feature consistency perspective

    Jiawei Zhang, Xiang Wang, Xiao Bai, Chen Wang, Lei Huang, Yimin Chen, Lin Gu, Jun Zhou, Tatsuya Harada, and Edwin R Hancock. Revisiting domain generalized stereo matching networks from a feature consistency perspective. InProceedings of the IEEE/CVF Conference on Computer Visi...

    2022

  56. [64]

    Graftnet: Towards domain generalized stereo match- ing with a broad-spectrum and task-oriented feature

    Biyang Liu, Huimin Yu, and Guodong Qi. Graftnet: Towards domain generalized stereo match- ing with a broad-spectrum and task-oriented feature. InProceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 13012–13021, 2022. 22

    2022

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.