Pith. sign in

REVIEW 4 major objections 5 minor 38 references

MultiFloodSynth: Multi-Annotated Flood Synthetic Dataset Generation

T0 review · 4 major / 5 minor · reviewed 2026-08-09 · deepseek-v4-flash

Pith's one-line read Adding 70,117 synthetic flood images to real training data improves flood-level detection for YOLOv10-B.

desk verdict MultiFloodSynth is a genuinely useful dataset contribution, but the evaluation protocol is too thin to confirm the claimed mAP gain. read the letter →

arxiv 2502.03966 v3 pith:7YRZHDL3 submitted 2025-02-06 cs.CV cs.AIcs.LG

classification cs.CVcs.AIcs.LG
keywords syntheticflooddatasetflood-leveldetectiondomainrandomizationmulti-annotationimage-to-3DgenerationurbanscenesimulationYOLOv10dataaugmentation
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper claims that a controllable virtual urban flood simulator can produce a large, consistently labeled synthetic dataset, MultiFloodSynth, that improves flood-level detection when used as training augmentation. The headline evidence is on YOLOv10-B: mixing the synthetic set with the real flood dataset raises mean average precision at 50% IoU from 56.66 to 58.61 and mAP50-95 from 40.64 to 42.71, while synthetic-only training is far weaker. The authors also report a Realistic Score of 93.17% relative to real data, which they interpret as near-real visual plausibility. If the improvement holds under a clean evaluation, the framework matters because it replaces costly, inconsistent hand labeling with automatic annotations for a disaster scenario that is hard to collect in the wild.

What carries the argument

The load-bearing mechanism is parameter-controllable scene composition in a 3D urban simulator: lighting, camera view, flood level, wave texture, roughness, opacity, and other settings are varied to generate diverse flood scenes. Domain randomization over these parameters, combined with image-to-3D generation for vehicles and city generation for base layouts, supplies the variety that makes the synthetic data useful for training. The simulator also auto-renders nine aligned annotation types, including segmentation, depth, normal maps, and 2D/3D bounding boxes, which avoids the inconsistent manual bounding boxes the paper attributes to real flood datasets.

What would settle it

Fix the real train/test split and random seeds, then train YOLOv10-B on Dreal alone and on Dreal plus Dsynth multiple times; if the reported mAP50 gain of 58.61 versus 56.66 does not reproduce, the central improvement claim is refuted. Separately, publishing the exact Realistic Score computation would allow checking whether 93.17% (with the real set normalized to 100) is a valid, reproducible value.

Watch

Extended reading notes

Core claim

The central discovery claimed here is that a synthetic flood dataset generated with editable scene parameters can augment, rather than replace, real flood imagery for object-localized flood-level detection. MultiFloodSynth contains 70,117 images spanning five flood levels and nine paired annotation types, and the paper's main result is that training on Dreal plus Dsynth outperforms training on Dreal alone: mAP50 rises from 56.66 to 58.61 for YOLOv10-B, and mAP50-95 rises from 40.64 to 42.71. The paper also reports that synthetic images reach 93.17% of the real dataset's Realistic Score, supporting the claim of on-par realism.

Load-bearing premise

The central claim collapses if the real dataset's five flood levels, defined by the percentage of a vehicle submerged, do not mean the same thing when applied to synthetic scenes; the paper does not provide a protocol for transferring or verifying that label definition.

Editorial extensions

If this is right

  • Training on Dreal plus Dsynth gives YOLOv10-B a mAP50 of 58.61 versus 56.66 for real data alone, so synthetic data acts as a useful supplement for flood-level detection.
  • Synthetic-only training performs far worse than real-only training (mAP50 6.94 versus 56.66 for YOLOv10-B), so MultiFloodSynth is not a replacement for real flood imagery.
  • The 70,117-image dataset with nine aligned annotation types provides consistently auto-labeled supervision for downstream tasks such as segmentation, depth estimation, and 3D box detection.
  • The reported Realistic Score of 93.17% implies that, under the cited metric, the synthetic frames are visually close to real flood scenes.

Reading between the lines

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

  • If the mAP gain is reproducible across random seeds and held-out test sets, the parameter-controlled simulator could be used to synthesize rare extreme flood levels that are scarce in real footage; the paper does not report per-level gains, so this remains an open test.
  • The 93.17% realism figure depends on a metric whose formal definition and normalization are not fully specified in the paper; a reproducible protocol would be needed to treat the realism claim as settled.
  • A similar controllable pipeline could plausibly extend to neighboring water-hazard domains such as storm surge or tsunami scenes, since the same wave, lighting, and camera controls apply.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper presents MultiFloodSynth, a synthetic urban flood dataset generated in NVIDIA Omniverse with 70,117 images across five flood-level classes and nine annotation types, including 2D/3D bounding boxes, segmentation maps, normal maps, and depth. The framework uses image-to-3D (Unique3D) for object generation, CityDreamer for city layouts, and a set of controllable parameters for urban and flood settings, with domain randomization to increase diversity. Experiments train YOLOv10-N and YOLOv10-B on real data only, synthetic data only, and their combination, reporting that mixing MultiFloodSynth with real data improves flood-level detection performance (e.g., YOLOv10-B mAP50 rises from 56.66 to 58.61 in Table 4b). Realism is assessed with a 'Realistic Score' borrowed from UrbanWorld (Shang et al. 2024), reporting 93.17% for the synthetic dataset relative to real data (Table 5).

Significance. If the dataset and its evaluation are fully validated, MultiFloodSynth would be a valuable contribution to flood hazard detection, offering a large, consistently annotated synthetic dataset that addresses a domain where real data collection is difficult and labels are often inconsistent. The multi-annotation design could support multiple downstream tasks beyond detection. However, the central empirical claims are not yet established because the evaluation protocol is under-specified and the realism metric is not reproducible from the manuscript alone. The potential is real, but the current evidence falls short of what is needed to support the headline conclusions.

major comments (4)
  1. [Comparison on Detection Performance (Table 4b)] The central claim that adding MultiFloodSynth to real data improves flood-level detection (mAP50 56.66 to 58.61 for YOLOv10-B) is not supported by the reported protocol: the paper does not state the real train/test split size, the number of random seeds or runs, or whether the same test split was used for all three training configurations. Given that the real dataset has only 2,000 frames (Table 2), the observed gain could easily be due to stochasticity or an accidental favorable split, so the reader cannot verify the main result.
  2. [Evaluation of MultiFloodSynth (Table 5)] The 'Realistic Score' is cited to Shang et al. 2024 but is not defined anywhere in this manuscript, and the normalization described only as 'We normalized the scores of Dsynth based on the scores of the real dataset' lacks the formula, the sampling protocol, and the code. Consequently, the reported 93.17% realism value cannot be independently recomputed or checked, making the 'on-par realism' claim unverifiable as presented.
  3. [Challenges of Real-World Flood Hazard Scenarios / Simulating Flood Wave] The real dataset's flood-level labels are defined by the percentage of a vehicle submerged (Wan et al. 2024), but the manuscript does not provide any bridging protocol to confirm that the synthetic flood-level labels correspond to the same physical definition. The large gap between synthetic-only training (mAP50 6.94 in Table 4b) and real-only training (56.66) also indicates a substantial domain shift that is never analyzed, so the interpretation of the mixed-training improvement as evidence of 'flood-level recognition' transfer is not yet justified.
  4. [Comparison on Detection Performance (Table 4)] The mixed-training improvement could stem from generic regularization or from simply adding more training data rather than from the specific content of MultiFloodSynth, since no comparison is made against standard data augmentation, an equal-size additional real-data baseline, or a synthetic dataset with different content. Such baselines are needed to attribute the observed mAP gains to the dataset's properties rather than to the extra training volume.
minor comments (5)
  1. [Abstract / Conclusions] The text contains several typos and ungrammatical phrases, e.g., 'our dataset demonstrate' (Abstract), 'alternating data requirements of real-world dataset' (Conclusions), and 'charactersize' (Conclusions), which should be corrected.
  2. [Table 2] The availability footnote '†Dataset will be available under the acceptance' is unclear about when and how the dataset will be released, and the Flood-Level row shows '4' for (Gao et al. 2024) with no explanation of what that number means relative to the other entries.
  3. [Related Works] The reference to Shang et al. 2024 appears twice in the first paragraph of Related Works, and the repeated citation should be consolidated.
  4. [Evaluation of MultiFloodSynth] When stating 'By randomly selecting 1K samples in each dataset, we average the score', the paper should specify whether the 1K selection was stratified by flood level or class and should report the variance of the averaged score rather than only the mean.
  5. [Figure 5] Figure 5 shows a point cloud in panel (h), but Table 2 does not list point clouds as an annotation type, so the annotation inventory is inconsistent between the table and the figure.

Circularity Check

0 steps flagged · score 2.0 of 10

No material circularity; the one minor self-citation is not load-bearing and the realism score is an external relative comparison.

full rationale

The paper's central claim is empirical: adding MultiFloodSynth to real training data improves flood-level detection mAP (Table 4). This is not derived from any fitted parameter or equation; the synthetic labels are set by simulator control parameters, and the evaluation is performed on the real test set. The 'Realistic Score' in Table 5 is borrowed from UrbanWorld (Shang et al. 2024), and normalizing the synthetic score by the real score (6.69/7.18 = 93.17%) is a relative comparison protocol rather than a definition that forces the claimed realism value. The only author self-citation is Jung et al. (2024) in the introduction, used as an example of a data-collection-intensive domain; the present framework and experiments do not rely on it. Missing experimental details, such as the real train/test split, number of runs, and the formal definition of the Realistic Score, are verifiability concerns rather than circularity. Accordingly, I find no step in which a predicted quantity reduces by construction to an input or to a load-bearing self-citation; the low score reflects only the incidental non-load-bearing self-citation.

Assumptions & free parameters 3 free parameters · 3 assumptions · 0 invented entities

The paper does not postulate new physical entities or novel mathematical constructs. Its load-bearing assumptions are the transferability of the flood-level taxonomy from real to synthetic data, the stability of single-run detection metrics, and the appropriateness of an undefined realism metric. The undisclosed ranges of simulator parameters function as implicit free parameters.

free parameters (3)
  • flood_level_definition = 5 levels described as percentage of vehicle submerged
    The five flood levels determine the target labels. They are inherited from Wan et al. 2024 and applied to synthetic scenes, but the exact submersion percentages are not stated in this paper.
  • realistic_score_threshold = 93.17 percent
    The reported realism score depends on the choice of 1,000 random samples per dataset and on the normalization against Dreal, both chosen by the authors.
  • Simulator control parameters = Not disclosed
    Table 1 lists the control parameters (e.g., lighting intensity, flood level, roughness, opacity) but no value ranges. The final dataset distribution depends on these unnamed choices.
assumptions (3)
  • domain assumption The real flood-level labels in Wan et al. 2024 are reliable enough to serve as the ground-truth definition for synthetic labels.
    The paper adopts Wan et al.'s label scheme (percentage of vehicle submerged) without questioning its consistency, although the paper's own motivation claims manual labels in real data are inconsistent.
  • domain assumption The YOLOv10 metrics (mAP50, mAP50-95) computed on a single run are stable enough to compare training configurations.
    No seeds or repeated runs are reported, so the reported improvements of 1 to 2 mAP points could be within run-to-run noise.
  • domain assumption The Realistic Score from Shang et al. 2024 is a valid measure of realism for flood imagery.
    The metric is borrowed from an urban city generation paper and applied without a formal definition or validation for flood scenes. Extreme values like 93.17 percent are assumed to be meaningful.

how reviews work

0 comments
Cite this review

Pith. "Pith review of MultiFloodSynth: Multi-Annotated Flood Synthetic Dataset Generation." pith.science (2026). https://pith.science/paper/7YRZHDL3

@misc{pith2026250203966,
  author       = {Pith},
  title        = {Pith review of: MultiFloodSynth: Multi-Annotated Flood Synthetic Dataset Generation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7YRZHDL3}},
  note         = {Machine review of arXiv:2502.03966}
}
read the original abstract

In this paper, we present synthetic data generation framework for flood hazard detection system. For high fidelity and quality, we characterize several real-world properties into virtual world and simulate the flood situation by controlling them. For the sake of efficiency, recent generative models in image-to-3D and urban city synthesis are leveraged to easily composite flood environments so that we avoid data bias due to the hand-crafted manner. Based on our framework, we build the flood synthetic dataset with 5 levels, dubbed MultiFloodSynth which contains rich annotation types like normal map, segmentation, 3D bounding box for a variety of downstream task. In experiments, our dataset demonstrate the enhanced performance of flood hazard detection with on-par realism compared with real dataset.

Figures

Figures reproduced from arXiv: 2502.03966 by the authors.

Figure 1
Figure 1. Overview of virtual flood scene composition and synthetic dataset generation pipeline. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Sample of real flood data (Wan et al. 2024) [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 4
Figure 4. Sample of our MultiFloodSynth. Simulating Flood Wave In flood hazard situation, flood simulating is pretty im￾portant factor which determine the flood level (Wan et al. 2024) and annotation-level. Some attributes (e.g., reflection, roughness, opacity, specular, texture) of flood object will di￾rectly affect to extract training feature by neural network. In this regards, we also simulate flood dynamics and visual ap￾… view at source ↗
Figures from the paper (1 more)
Figure 6
Figure 6. Figure 6: Visualization of EigenCAM for explainability. A [PITH_FULL_IMAGE:figures/full_fig_p004_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

38 extracted references · 28 canonical work pages

  1. [1]

    , " * write output.state after.block = add.period write newline

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.a...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

  3. [3]

    A.; Klein, J.; Liu, D.; Pa ubicki, W.; Pirk, S.; and Michels, D

    Amador Herrera, J. A.; Klein, J.; Liu, D.; Pa ubicki, W.; Pirk, S.; and Michels, D. L. 2024. Cyclogenesis: Simulating Hurricanes and Tornadoes. ACM Transactions on Graphics (TOG), 43(4): 1--16

  4. [4]

    P.; Schindler, K.; and Wegner, J

    Chaudhary, P.; D’Aronco, S.; Leit \ a o, J. P.; Schindler, K.; and Wegner, J. D. 2020. Water level prediction from social media images with a multi-task ranking approach. ISPRS Journal of Photogrammetry and Remote Sensing, 167: 252--262

  5. [5]

    Delussu, R.; Putzu, L.; and Fumera, G. 2024. Synthetic Data for Video Surveillance Applications of Computer Vision: A Review. International Journal of Computer Vision, 1--37

  6. [6]

    E.; Dhakad, S.; Vishwakarma, S.; Wang, C.; Jhang, Y.-C.; Chociej, M.; Crespi, A.; Thaman, A.; and Ganguly, S

    Ebadi, S. E.; Dhakad, S.; Vishwakarma, S.; Wang, C.; Jhang, Y.-C.; Chociej, M.; Crespi, A.; Thaman, A.; and Ganguly, S. 2022. PSP-HDRI + : A Synthetic Dataset Generator for Pre-Training of Human-Centric Computer Vision Models. arXiv:2207.05025

  7. [7]

    Gao, K.; Yang, Z.; Gao, X.; Shao, W.; Wei, H.; and Xu, T. 2024. Measuring urban waterlogging depths from video images based on reference objects. Journal of Flood Risk Management, 17(1): e12948

  8. [8]

    J.; Gnanapragasam, D.; Golemo, F.; Herrmann, C.; Kipf, T.; Kundu, A.; Lagun, D.; Laradji, I

    Greff, K.; Belletti, F.; Beyer, L.; Doersch, C.; Du, Y.; Duckworth, D.; Fleet, D. J.; Gnanapragasam, D.; Golemo, F.; Herrmann, C.; Kipf, T.; Kundu, A.; Lagun, D.; Laradji, I. H.; Liu, H.-T.; Meyer, H.; Miao, Y.; Nowrouzezahrai, D.; Oztireli, C.; Pot, E.; Radwan, N.; Rebain, D.; Sabour, S.; Sajjadi, M. S. M.; Sela, M.; Sitzmann, V.; Stone, A.; Sun, D.; Vor...

Show all 38 references
  1. [9]

    Hamza, A.; Lojo, A.; N \'u \ n ez-Marcos, A.; and Atutxa, A. 2024. Ali-AUG: Innovative Approaches to Labeled Data Augmentation using One-Step Diffusion Model. arXiv:2410.18678

  2. [10]

    Hao, S.; Han, W.; Jiang, T.; Li, Y.; Wu, H.; Zhong, C.; Zhou, Z.; and Tang, H. 2024. Synthetic data in AI: Challenges, applications, and ethical implications. arXiv:2401.01629

  3. [11]

    Hong, Z.; Hamdan, E.; Zhao, Y.; Ye, T.; Pan, H.; and Cetin, A. E. 2024. Wildfire detection via transfer learning: a survey. Signal, Image and Video Processing, 18(1): 207--214

  4. [12]

    Hummel, M.; and van Kooten, K. 2019. Leveraging nvidia omniverse for in situ visualization. In High Performance Computing: ISC High Performance 2019 International Workshops, Frankfurt, Germany, June 16-20, 2019, Revised Selected Papers 34, 634--642. Springer

  5. [13]

    Z.; Mahmood, A.; and Nandakumar, K

    Islam, K.; Zaheer, M. Z.; Mahmood, A.; and Nandakumar, K. 2024. DiffuseMix: Label-Preserving Data Augmentation with Diffusion Models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 27621--27630

  6. [14]

    U.; and Seo, S

    Jung, Y.; Byun, S.; Kim, B.; Amin, S. U.; and Seo, S. 2024. Harnessing synthetic data for enhanced detection of Pine Wilt Disease: An image classification approach. Computers and Electronics in Agriculture, 218: 108690

  7. [15]

    Karanjit, R.; Pally, R.; and Samadi, S. 2023. FloodIMG: flood image DataBase system. Data in brief, 48: 109164

  8. [16]

    Khullar, D.; Sokhandan, N.; Kulkarni, N.; and Shah, Y. 2023. Synthetic data generation for scarce road scene detection scenarios

  9. [17]

    Kim, H.-C.; Lam, H.-K.; Lee, S.-H.; and Ok, S.-Y. 2024. Early Fire Detection System by Using Automatic Synthetic Dataset Generation Model Based on Digital Twins. Applied Sciences, 14(5): 1801

  10. [18]

    L.; Pirk, S.; and Palubicki, W

    Kokosza, A.; Wrede, H.; Gonzalez Esparza, D.; Makowski, M.; Liu, D.; Michels, D. L.; Pirk, S.; and Palubicki, W. 2024. Scintilla: Simulating Combustible Vegetation for Wildfires. ACM Transactions on Graphics (TOG), 43(4): 1--21

  11. [19]

    Lee, K.; Shin, U.; and Lee, B.-U. 2024. Learning to Control Camera Exposure via Reinforcement Learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2975--2983

  12. [20]

    Lin, C.-H.; Gao, J.; Tang, L.; Takikawa, T.; Zeng, X.; Huang, X.; Kreis, K.; Fidler, S.; Liu, M.-Y.; and Lin, T.-Y. 2023. Magic3d: High-resolution text-to-3d content creation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 300--309

  13. [21]

    Lo, S.-W.; Wu, J.-H.; Chang, J.-Y.; Tseng, C.-H.; Lin, M.-W.; and Lin, F.-P. 2021. Deep sensing of urban waterlogging. IEEE Access, 9: 127185--127203

  14. [22]

    L.; Singh, R.; Guo, Y.; Mazhar, H.; et al

    Mittal, M.; Yu, C.; Yu, Q.; Liu, J.; Rudin, N.; Hoeller, D.; Yuan, J. L.; Singh, R.; Guo, Y.; Mazhar, H.; et al. 2023. Orbit: A unified simulation framework for interactive robot learning environments. IEEE Robotics and Automation Letters, 8(6): 3740--3747

  15. [23]

    B.; and Yeasin, M

    Muhammad, M. B.; and Yeasin, M. 2020. Eigen-cam: Class activation map using principal components. In 2020 international joint conference on neural networks (IJCNN), 1--7. IEEE

  16. [24]

    Pally, R.; and Samadi, S. 2022. Application of image processing and convolutional neural networks for flood image classification and semantic segmentation. Environmental modelling & software, 148: 105285

  17. [25]

    Rawal, P.; Sompura, M.; and Hintze, W. 2023. Synthetic data generation for bridging Sim2Real gap in a production environment. arXiv:2311.11039

  18. [26]

    R.; AlHaija, H

    Richter, S. R.; AlHaija, H. A.; and Koltun, V. 2022. Enhancing photorealism enhancement. IEEE Transactions on Pattern Analysis and Machine Intelligence, 45(2): 1700--1715

  19. [27]

    C.; Kleinbeck, C.; Yang, S

    Schieber, H.; Demir, K. C.; Kleinbeck, C.; Yang, S. H.; and Roth, D. 2024. Indoor synthetic data generation: A systematic review. Computer Vision and Image Understanding, 103907

  20. [28]

    Shang, Y.; Lin, Y.; Zheng, Y.; Fan, H.; Ding, J.; Feng, J.; Chen, J.; Tian, L.; and Li, Y. 2024. UrbanWorld: An Urban World Model for 3D City Generation. arXiv:2407.11965

  21. [29]

    Valvano, G.; Agostino, A.; De Magistris, G.; Graziano, A.; and Veneri, G. 2024. Controllable Image Synthesis of Industrial Data using Stable Diffusion. In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, 5354--5363

  22. [30]

    Wan, J.; Qin, Y.; Shen, Y.; Yang, T.; Yan, X.; Zhang, S.; Yang, G.; Xue, F.; and Wang, Q. J. 2024. Automatic detection of urban flood level with YOLOv8 using flooded vehicle dataset. Journal of Hydrology, 639: 131625

  23. [31]

    Wang, A.; Chen, H.; Liu, L.; Chen, K.; Lin, Z.; Han, J.; and Ding, G. 2024 a . Yolov10: Real-time end-to-end object detection. arXiv:2405.14458

  24. [32]

    Wang, Z.; Draghi, B.; Rotalinti, Y.; Lunn, D.; and Myles, P. 2024 b . High-fidelity synthetic data applications for data augmentation

  25. [33]

    Wu, K.; Liu, F.; Cai, Z.; Yan, R.; Wang, H.; Hu, Y.; Duan, Y.; and Ma, K. 2024 a . Unique3D: High-Quality and Efficient 3D Mesh Generation from a Single Image. arXiv:2405.20343

  26. [34]

    Wu, L.; Liu, Y.; Zhang, J.; Zhang, B.; Wang, Z.; Tong, J.; Li, M.; and Zhang, A. 2024 b . Identification of flood depth levels in urban waterlogging disaster caused by rainstorm using a CBAM-improved ResNet50. Expert Systems with Applications, 124382

  27. [35]

    Xie, H.; Chen, Z.; Hong, F.; and Liu, Z. 2024. Citydreamer: Compositional generative model of unbounded 3d cities. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 9666--9675

  28. [36]

    Zhang, S.; Zhou, M.; Wang, Y.; Luo, C.; Wang, R.; Li, Y.; Yin, X.; Zhang, Z.; and Peng, J. 2024. CityX: Controllable Procedural Content Generation for Unbounded 3D Cities. arXiv:2407.17572

  29. [37]

    Zhong, P.; Liu, Y.; Zheng, H.; and Zhao, J. 2024. Detection of urban flood inundation from traffic images using deep learning methods. Water Resources Management, 38(1): 287--301

  30. [38]

    Zhu, J.; Li, S.; Liu, Y.; Huang, P.; Shan, J.; Ma, H.; and Yuan, J. 2024. ODGEN: Domain-specific Object Detection Data Generation with Diffusion Models. arXiv:2405.15199

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.