pith. machine review for the scientific record. sign in

arxiv: 2604.02396 · v1 · submitted 2026-04-02 · 💻 cs.CV · cs.AI

Recognition: 2 theorem links

· Lean Theorem

Environment-Aware Channel Prediction for Vehicular Communications: A Multimodal Visual Feature Fusion Framework

Xuejian Zhang , Ruisi He , Minseok Kim , Inocent Calist , Mi Yang , Ziyi Qi
Authors on Pith no claims yet

Pith reviewed 2026-05-13 21:53 UTC · model grok-4.3

classification 💻 cs.CV cs.AI
keywords environment-aware channel predictionvehicular communicationsmultimodal feature fusionsemantic segmentationdepth estimationangular power spectrumV2I communications
0
0 comments X

The pith

Fusing semantic and depth features from vehicle cameras predicts radio channel metrics like path loss and angular spreads in urban settings.

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

The paper develops a multimodal framework that takes panoramic RGB images and GPS data from vehicles to forecast radio propagation parameters ahead of transmission. It runs images through separate branches for semantic segmentation and depth estimation, merges those outputs with position information using attention-based gating, and produces joint estimates of path loss, delay spread, azimuth spreads, and the full angular power spectrum. Traditional models struggle with accuracy and adaptability in changing environments, so this approach tests whether visual environmental cues can supply the missing priors for reliable prediction. Results on real synchronized urban measurements show low errors across the metrics, pointing toward more responsive channel use in high-mobility links.

Core claim

The environment-aware framework extracts semantic, depth, and position features via a three-branch network from vehicle-side panoramic RGB images and GPS, fuses them adaptively through a squeeze-excitation attention module, and applies a dedicated regression head with composite loss to jointly predict path loss, delay spread, azimuth spread of arrival, azimuth spread of departure, and 360-dimensional angular power spectrum, reaching 3.26 dB RMSE for path loss and 0.9342 mean cosine similarity for the spectrum on urban V2I data.

What carries the argument

Three-branch architecture that extracts semantic segmentation, depth estimation, and position features from RGB images, then performs adaptive fusion via squeeze-excitation attention gating before regression to channel parameters.

If this is right

  • Path loss can be predicted to 3.26 dB RMSE using only onboard visual and position data.
  • Delay spread reaches 37.66 ns RMSE accuracy without direct channel sounding.
  • Azimuth spreads of arrival and departure achieve roughly 5 degree RMSE.
  • Full angular power spectrum maintains mean cosine similarity above 0.93.
  • Joint multi-metric prediction supports lower-latency adaptation in vehicular links.

Where Pith is reading between the lines

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

  • The same visual pipeline could reduce pilot overhead by supplying priors that cut the frequency of explicit channel measurements.
  • Extending the fusion module to include LiDAR or radar point clouds might tighten predictions in low-visibility conditions.
  • Retraining on mixed indoor-outdoor datasets could test whether the same structure transfers beyond pure urban V2I.

Load-bearing premise

Visual semantic and depth features from RGB images supply enough information about scattering, blockage, and other radio effects to generalize across different urban scenes.

What would settle it

A sharp rise in prediction error when the trained model is evaluated on synchronized measurements collected in a new city whose building layouts and materials differ substantially from the training set.

Figures

Figures reproduced from arXiv: 2604.02396 by Inocent Calist, Minseok Kim, Mi Yang, Ruisi He, Xuejian Zhang, Ziyi Qi.

Figure 1
Figure 1. Figure 1: The proposed channel prediction model architecture based on visual sensing and multimodal feature fusion. [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Regression head architecture specifically designed for APS prediction. [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Measurement system. d) Relative Total Power Constraint Loss (LRTP): To ensure that the global total power of Pbm is consistent with P⋆ m, LRTP is designed to avoid loss imbalance caused by differences in total power across different samples: LRTP = 1 M X M m=1 [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 5
Figure 5. Figure 5: The processing pipeline are described in detail below. [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 4
Figure 4. Figure 4: Measurement trajectories along three different Areas in urban scenarios. (a) [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Block diagram of the dataset generation pipeline. [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Four representative types of panoramic images: (a) raw panoramic [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 9
Figure 9. Figure 9: Comparison of the raw image and images with dynamic scatterers [PITH_FULL_IMAGE:figures/full_fig_p010_9.png] view at source ↗
Figure 8
Figure 8. Figure 8: Results of the impact of different modal inputs on prediction accuracy: [PITH_FULL_IMAGE:figures/full_fig_p010_8.png] view at source ↗
Figure 12
Figure 12. Figure 12: Results of the impact of different backbone networks on prediction [PITH_FULL_IMAGE:figures/full_fig_p011_12.png] view at source ↗
Figure 10
Figure 10. Figure 10: Results of the impact of eliminating dynamic scatterers on prediction [PITH_FULL_IMAGE:figures/full_fig_p011_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Comparison of training error versus epoch with and without dynamic [PITH_FULL_IMAGE:figures/full_fig_p011_11.png] view at source ↗
Figure 14
Figure 14. Figure 14: Probability density distribution of cosine similarity of all samples in [PITH_FULL_IMAGE:figures/full_fig_p012_14.png] view at source ↗
Figure 13
Figure 13. Figure 13: Examples of APS prediction results. D. APS Predictive Performance Verification APS is a 360-dimensional vector that describes the angu￾lar distribution of multipath energy. Its prediction accuracy therefore directly reflects the ability of model to capture fine grained angular domain structure. The experiment still takes semantic segmentation image, depth image, and GPS data as trimodal input. Data from A… view at source ↗
read the original abstract

The deep integration of communication with intelligence and sensing, as a defining vision of 6G, renders environment-aware channel prediction a key enabling technology. As a representative 6G application, vehicular communications require accurate and forward-looking channel prediction under stringent reliability, latency, and adaptability demands. Traditional empirical and deterministic models remain limited in balancing accuracy, generalization, and deployability, while the growing availability of onboard and roadside sensing devices offers a promising source of environmental priors. This paper proposes an environment-aware channel prediction framework based on multimodal visual feature fusion. Using GPS data and vehicle-side panoramic RGB images, together with semantic segmentation and depth estimation, the framework extracts semantic, depth, and position features through a three-branch architecture and performs adaptive multimodal fusion via a squeeze-excitation attention gating module. For 360-dimensional angular power spectrum (APS) prediction, a dedicated regression head and a composite multi-constraint loss are further designed. As a result, joint prediction of path loss (PL), delay spread (DS), azimuth spread of arrival (ASA), azimuth spread of departure (ASD), and APS is achieved. Experiments on a synchronized urban V2I measurement dataset yield the best root mean square error (RMSE) of 3.26 dB for PL, RMSEs of 37.66 ns, 5.05 degrees, and 5.08 degrees for DS, ASA, and ASD, respectively, and mean/median APS cosine similarities of 0.9342/0.9571, demonstrating strong accuracy, generalization, and practical potential for intelligent channel prediction in 6G vehicular communications.

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

3 major / 2 minor

Summary. The paper proposes an environment-aware channel prediction framework for vehicular communications that fuses multimodal visual features from panoramic RGB images (via semantic segmentation and depth estimation) with GPS data using a three-branch architecture and squeeze-excitation attention for adaptive fusion. It predicts path loss (PL), delay spread (DS), azimuth spreads (ASA, ASD), and 360-dimensional angular power spectrum (APS) with a dedicated regression head and multi-constraint loss. On a synchronized urban V2I measurement dataset, it reports RMSE of 3.26 dB for PL, 37.66 ns for DS, 5.05° for ASA, 5.08° for ASD, and mean/median APS cosine similarities of 0.9342/0.9571.

Significance. If the reported performance holds under rigorous validation, this work could advance 6G vehicular communications by showing how onboard sensing enables more accurate forward-looking channel prediction, potentially outperforming traditional empirical models in dynamic environments. The multimodal fusion approach aligns with the vision of integrating communication, intelligence, and sensing.

major comments (3)
  1. [Experiments] Experimental evaluation: No ablation studies are reported that isolate the contribution of the visual semantic and depth branches (e.g., by removing them and retraining on GPS alone), so it is unclear whether the reported RMSE and cosine similarity gains are attributable to multimodal fusion or to dataset-specific correlations.
  2. [Experiments] Dataset and evaluation protocol: The results are obtained on a single synchronized urban V2I measurement campaign with no cross-scenario splits (different street layouts, building materials, or cities) or held-out test environments, leaving the generalization claims unsupported by evidence.
  3. [Experiments] Statistical reporting: The abstract and results provide point estimates for RMSE and APS similarities but omit training/validation split details, number of samples, error bars, or statistical significance tests, which are required to substantiate the central performance claims.
minor comments (2)
  1. [Methodology] The abstract mentions a 'composite multi-constraint loss' for APS regression but does not specify the individual loss terms or their weighting; this should be clarified in the methodology section.
  2. [Figures] Figure captions and axis labels for any performance plots should explicitly state the number of test samples and whether results are averaged over multiple runs.

Simulated Author's Rebuttal

3 responses · 1 unresolved

We thank the referee for the constructive comments, which help improve the rigor of our work. We address each major comment point-by-point below, indicating planned revisions where feasible.

read point-by-point responses

  1. Referee: [Experiments] Experimental evaluation: No ablation studies are reported that isolate the contribution of the visual semantic and depth branches (e.g., by removing them and retraining on GPS alone), so it is unclear whether the reported RMSE and cosine similarity gains are attributable to multimodal fusion or to dataset-specific correlations.

    Authors: We agree that ablation studies are necessary to isolate modality contributions. The revised manuscript will include new ablation experiments: full model vs. GPS-only, semantic-only, depth-only, and pairwise combinations, with quantitative comparisons of RMSE and APS similarity to demonstrate the gains from multimodal fusion. revision: yes

  2. Referee: [Experiments] Dataset and evaluation protocol: The results are obtained on a single synchronized urban V2I measurement campaign with no cross-scenario splits (different street layouts, building materials, or cities) or held-out test environments, leaving the generalization claims unsupported by evidence.

    Authors: Our results are from one comprehensive urban campaign. We will add intra-dataset splits (e.g., by route segments or time-of-day) for better internal validation and explicitly discuss limitations on broader generalization. Cross-city experiments are not possible without new measurements. revision: partial

  3. Referee: [Experiments] Statistical reporting: The abstract and results provide point estimates for RMSE and APS similarities but omit training/validation split details, number of samples, error bars, or statistical significance tests, which are required to substantiate the central performance claims.

    Authors: We will revise the results section to report dataset size, exact train/validation/test splits, standard deviations from multiple runs, and statistical significance tests (e.g., t-tests) to support the performance claims. revision: yes

standing simulated objections not resolved
  • Cross-scenario validation on datasets from different cities or environments, as this requires new synchronized measurement campaigns beyond the scope of the current study.

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper presents a standard supervised multimodal neural network for regressing channel statistics (PL, DS, ASA, ASD, APS) from fused visual semantic/depth features and GPS inputs. The architecture, attention gating, and composite loss are conventional design choices; the reported RMSE and cosine similarity values are model outputs evaluated on held-out measurement data rather than quantities defined by construction from the same fitted parameters. No equations reduce predictions to inputs, no self-citation chain supplies a uniqueness theorem, and no ansatz is smuggled via prior work. The derivation is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review provides no explicit free parameters, axioms, or invented entities; the framework implicitly assumes that image-derived features correlate with radio propagation without stating the functional form of that mapping.

pith-pipeline@v0.9.0 · 5607 in / 1191 out tokens · 30379 ms · 2026-05-13T21:53:38.321680+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

34 extracted references · 34 canonical work pages · 1 internal anchor

  1. [1]

    Environment- aware path loss prediction using panoramic images for vehicular com- munications,

    X. Zhang, M. Kim, I. Calist, R. He, M. Yang, and Z. Qi, “Environment- aware path loss prediction using panoramic images for vehicular com- munications,” inProc. IEEE Int. Conf. Commun. (ICC), accepted, 2026, pp. 1–6

    work page 2026

  2. [2]

    Framework and overall objectives of the future development of IMT for 2030 and beyond,

    ITU-R, “Framework and overall objectives of the future development of IMT for 2030 and beyond,”DRAFT NEW RECOMMENDATION, Jun. 2023

    work page 2030

  3. [3]

    A general channel model for integrated sensing and communication scenarios,

    Z. Zhanget al., “A general channel model for integrated sensing and communication scenarios,”IEEE Commun. Mag., vol. 61, no. 5, pp. 68–74, 2022

    work page 2022

  4. [4]

    COST CA20120 INTERACT framework of artificial intelligence-based channel modeling,

    R. He, N. D. Cicco, B. Ai, M. Yang, Y . Miao, and M. Boban, “COST CA20120 INTERACT framework of artificial intelligence-based channel modeling,”IEEE Wirel. Commun., vol. 32, no. 4, pp. 200–207, 2025

    work page 2025

  5. [5]

    Propagation channels of 5G millimeter-wave vehicle-to- vehicle communications: Recent advances and future challenges,

    R. Heet al., “Propagation channels of 5G millimeter-wave vehicle-to- vehicle communications: Recent advances and future challenges,”IEEE Veh. Technol. Mag., vol. 15, no. 1, pp. 16–26, 2020

    work page 2020

  6. [6]

    Artificial intelligence empowered channel prediction: A new paradigm for propagation channel modeling,

    R. He, M. Yang, Z. Zhang, B. Ai, and Z. Zhong, “Artificial intelligence empowered channel prediction: A new paradigm for propagation channel modeling,”arXiv preprint arXiv:2601.09205, 2026

    work page arXiv 2026

  7. [7]

    Artificial intelligence enabled radio propagation for communications—Part I: Channel characterization and antenna-channel optimization,

    C. Huanget al., “Artificial intelligence enabled radio propagation for communications—Part I: Channel characterization and antenna-channel optimization,”IEEE Trans. Antennas Propag., vol. 70, no. 6, pp. 3939– 3954, 2022

    work page 2022

  8. [8]

    Multi-modal intelligent channel modeling: A new modeling paradigm via synesthesia of machines,

    L. Bai, Z. Huang, M. Sun, X. Cheng, and L. Cui, “Multi-modal intelligent channel modeling: A new modeling paradigm via synesthesia of machines,”IEEE Commun. Surveys Tuts., vol. 28, pp. 2612–2649, 2026

    work page 2026

  9. [9]

    Applying deep-learning-based com- puter vision to wireless communications: Methodologies, opportunities, and challenges,

    Y . Tian, G. Pan, and M.-S. Alouini, “Applying deep-learning-based com- puter vision to wireless communications: Methodologies, opportunities, and challenges,”IEEE Open J. Commun. Society, vol. 2, pp. 132–143, 2021

    work page 2021

  10. [10]

    When wireless communications meet computer vision in beyond 5G,

    T. Nishio, Y . Koda, J. Park, M. Bennis, and K. Doppler, “When wireless communications meet computer vision in beyond 5G,”IEEE Commun. Standards Mag., vol. 5, no. 2, pp. 76–83, 2021

    work page 2021

  11. [11]

    Vision-aided 6G wireless communications: Blockage prediction and proactive handoff,

    G. Charan, M. Alrabeiah, and A. Alkhateeb, “Vision-aided 6G wireless communications: Blockage prediction and proactive handoff,”IEEE Trans. Veh. Technol., vol. 70, no. 10, pp. 10 193–10 208, 2021

    work page 2021

  12. [12]

    Environment semantics aided wireless communications: A case study of mmWave beam pre- diction and blockage prediction,

    Y . Yang, F. Gao, X. Tao, G. Liu, and C. Pan, “Environment semantics aided wireless communications: A case study of mmWave beam pre- diction and blockage prediction,”IEEE J. Sel. Areas Commun., vol. 41, no. 7, pp. 2025–2040, 2023

    work page 2025

  13. [13]

    Environment semantic com- munication: Enabling distributed sensing aided networks,

    S. Imran, G. Charan, and A. Alkhateeb, “Environment semantic com- munication: Enabling distributed sensing aided networks,”IEEE Open Journal of the Communications Society, vol. 5, pp. 7767–7786, 2024

    work page 2024

  14. [14]

    Camera based mmWave beam prediction: Towards multi-candidate real-world scenarios,

    G. Charan, M. Alrabeiah, T. Osman, and A. Alkhateeb, “Camera based mmWave beam prediction: Towards multi-candidate real-world scenarios,”IEEE Trans. Veh. Technol., vol. 74, no. 4, pp. 5897–5913, 2025

    work page 2025

  15. [15]

    Deepsense 6G: A large-scale real-world multi-modal sensing and communication dataset,

    A. Alkhateeb, G. Charan, T. Osman, A. Hredzak, J. Morais, U. Demirhan, and N. Srinivas, “Deepsense 6G: A large-scale real-world multi-modal sensing and communication dataset,”IEEE Commun. Mag., vol. 61, no. 9, pp. 122–128, 2023

    work page 2023

  16. [16]

    Environment sensing- aided beam prediction with transfer learning for smart factory,

    Y . Feng, C. Zhao, F. Gao, Y . Zhang, and S. Ma, “Environment sensing- aided beam prediction with transfer learning for smart factory,”IEEE Trans. Wireless Commun., vol. 24, no. 1, pp. 676–690, 2025

    work page 2025

  17. [17]

    Proactive received power prediction using machine learning and depth images for mmWave networks,

    T. Nishio, H. Okamoto, K. Nakashima, Y . Koda, K. Yamamoto, M. Morikura, Y . Asai, and R. Miyatake, “Proactive received power prediction using machine learning and depth images for mmWave networks,”IEEE J. Sel. Areas Commun., vol. 37, no. 11, pp. 2413–2427, 2019

    work page 2019

  18. [18]

    Vision aided channel prediction for vehicular communications: A case study of received power prediction using RGB images,

    X. Zhang, R. He, M. Yang, Z. Zhang, Z. Qi, and B. Ai, “Vision aided channel prediction for vehicular communications: A case study of received power prediction using RGB images,”IEEE Trans. Veh. Technol., vol. 74, no. 11, pp. 17 531–17 544, 2025

    work page 2025

  19. [19]

    Multi- modal environmental information sensing based path loss prediction for V2I communications,

    K. Wang, L. Yu, J. Zhang, Y . Tian, E. Guo, and G. Liu, “Multi- modal environmental information sensing based path loss prediction for V2I communications,” inProc. IEEE 101st Veh. Technol. Conf. (VTC- Spring), 2025, pp. 1–5

    work page 2025

  20. [20]

    An intelligent path loss prediction approach based on integrated sensing and communi- cations for future vehicular networks,

    Z. Wei, B. Mao, H. Guo, Y . Xun, J. Liu, and N. Kato, “An intelligent path loss prediction approach based on integrated sensing and communi- cations for future vehicular networks,”IEEE Open J. Commun. Society, vol. 5, pp. 170–180, 2024

    work page 2024

  21. [21]

    Path loss pre- diction for vehicle-to-infrastructure communications via synesthesia of machines (SoM),

    M. Lu, L. Bai, Z. Huang, M. Yang, and X. Cheng, “Path loss pre- diction for vehicle-to-infrastructure communications via synesthesia of machines (SoM),”Radio Sci., vol. 60, no. 6, pp. 1–15, 2025

    work page 2025

  22. [22]

    Multi-modal sensing data- based real-time path loss prediction for 6G UA V-to-ground communi- cations,

    M. Sun, L. Bai, Z. Huang, and X. Cheng, “Multi-modal sensing data- based real-time path loss prediction for 6G UA V-to-ground communi- cations,”IEEE Wireless Commun. Lett., vol. 13, no. 9, pp. 2462–2466, 2024

    work page 2024

  23. [23]

    Vision-aided channel prediction based on image segmentation at street intersection scenarios,

    X. Zhang, R. He, M. Yang, Z. Qi, Z. Zhang, B. Ai, and Z. Zhong, “Vision-aided channel prediction based on image segmentation at street intersection scenarios,”IEEE Trans. on Cogn. Commun. Netw., vol. 12, pp. 1678–1693, 2026

    work page 2026

  24. [24]

    A multimodal predictive channel model based on dual-camera images for IIoT communications,

    S. Zhou, Y . Liu, R. Wang, Z. Li, Z. Xin, J. Huang, and J. Bian, “A multimodal predictive channel model based on dual-camera images for IIoT communications,”IEEE Internet Things J., vol. 12, no. 12, pp. 20 530–20 543, 2025

    work page 2025

  25. [25]

    Multimodal fusion-based channel prediction and characterization for mmWave UA V A2G communications,

    Z. Xin, Y . Liu, J. Xing, J. Huang, J. Bian, Z. Bai, and C. Wang, “Multimodal fusion-based channel prediction and characterization for mmWave UA V A2G communications,”IEEE Trans. Commun., vol. 74, pp. 5089–5104, 2026

    work page 2026

  26. [26]

    A cluster-based predictive channel modeling for mmwave communications via deep transfer learning: A multimodal data-driven approach,

    B. Yin, Y . Miao, A. Bodi, R. Caromi, J. Senic, C. Gentile, W. Joseph, and M. Deruyck, “A cluster-based predictive channel modeling for mmwave communications via deep transfer learning: A multimodal data-driven approach,”IEEE Trans. Veh. Technol., pp. 1–16, 2026

    work page 2026

  27. [27]

    M3SC: A generic dataset for mixed multi-modal (MMM) sensing and communication integration,

    X. Cheng, Z. Huang, L. Bai, H. Zhang, M. Sun, B. Liu, S. Li, J. Zhang, and M. Lee, “M3SC: A generic dataset for mixed multi-modal (MMM) sensing and communication integration,”China Commun., vol. 20, no. 11, pp. 13–29, 2023

    work page 2023

  28. [28]

    Deep residual learning for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” inProc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), June 2016

    work page 2016

  29. [29]

    Novel scalable MIMO chan- nel sounding technique and measurement accuracy evaluation with transceiver impairments,

    M. Kim, J.-i. Takada, and Y . Konishi, “Novel scalable MIMO chan- nel sounding technique and measurement accuracy evaluation with transceiver impairments,”IEEE Trans. Instrum. Meas., vol. 61, no. 12, pp. 3185–3197, 2012

    work page 2012

  30. [30]

    Bridging FR1 to FR3: Urban Channel Parameterization Anchored at 4.85 GHz and Literature-Referenced Cross-Band Trends

    I. Calist and M. Kim, “Bridging FR1 to FR3: Frequency-continuous urban macro/microcellular channel parameterization anchored at 4.85 GHz,”arXiv preprint arXiv:2512.00707, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  31. [31]

    Masked-attention mask transformer for universal image segmentation,

    B. Cheng, I. Misra, A. G. Schwing, A. Kirillov, and R. Girdhar, “Masked-attention mask transformer for universal image segmentation,” inProc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), 2022, pp. 1280–1289

    work page 2022

  32. [32]

    Depth anything v2,

    L. Yang, B. Kang, Z. Huang, Z. Zhao, X. Xu, J. Feng, and H. Zhao, “Depth anything v2,” inProc. Int. Conf. Neural Inf. Process. Syst., 2024, pp. 21 875–21 911

    work page 2024

  33. [33]

    The cityscapes dataset for semantic urban scene understanding,

    M. Cordtset al., “The cityscapes dataset for semantic urban scene understanding,” inProc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), 2016

    work page 2016

  34. [34]

    Imagenet large scale visual recognition chal- lenge,

    O. Russakovskyet al., “Imagenet large scale visual recognition chal- lenge,”Int. J. Comput. Vis., vol. 115, pp. 211–252, 2015

    work page 2015