arxiv: 2604.21801 · v1 · submitted 2026-04-23 · 💻 cs.CV · cs.AI

Recognition: unknown

SyMTRS: Benchmark Multi-Task Synthetic Dataset for Depth, Domain Adaptation and Super-Resolution in Aerial Imagery

Safouane El Ghazouali , Nicola Venturi , Michael Rueegsegger , Umberto Michelucci

Authors on Pith no claims yet

Pith reviewed 2026-05-09 22:43 UTC · model grok-4.3

classification 💻 cs.CV cs.AI

keywords synthetic datasetaerial imagerydepth estimationdomain adaptationsuper-resolutionremote sensingmulti-task benchmark

0 comments

The pith

SyMTRS supplies a single synthetic aerial dataset with pixel-perfect depth maps, night-time pairs, and multi-scale low-resolution images.

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

The paper introduces SyMTRS, a large-scale synthetic dataset generated from a high-fidelity urban simulation pipeline. It supplies 2048 by 2048 RGB aerial imagery together with aligned pixel-perfect depth, night-time counterparts, and downsampled versions at 2x, 4x, and 8x scales. This unified collection lets researchers train and test models on monocular depth estimation, day-to-night domain adaptation, and super-resolution within the same scenes and with perfect ground truth. The authors position the dataset as a way to overcome the cost and scarcity of real annotated remote-sensing data for these three tasks.

Core claim

SyMTRS is a multi-task synthetic benchmark that supplies high-resolution RGB aerial imagery, pixel-perfect depth maps, night-time domain-shift pairs, and aligned low-resolution variants at x2, x4, and x8 scales, all produced by a single high-fidelity urban simulation pipeline.

What carries the argument

The high-fidelity urban simulation pipeline that generates geometrically consistent, multi-domain aerial imagery with perfect depth and scale annotations.

Load-bearing premise

Imagery produced by the simulation pipeline has statistical properties and variations close enough to real aerial remote-sensing data that models trained on it will transfer.

What would settle it

Train a monocular depth model on SyMTRS and evaluate it on a real-world aerial depth dataset; performance substantially below that of models trained on real data would falsify the transfer assumption.

Figures

Figures reproduced from arXiv: 2604.21801 by Michael Rueegsegger, Nicola Venturi, Safouane El Ghazouali, Umberto Michelucci.

**Figure 2.** Figure 2: Sample representation of the dataset components which include: Raw RGB high [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗

**Figure 3.** Figure 3: SR quantitative comparison aggregated from the test split for [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗

**Figure 4.** Figure 4: SR quantitative comparison aggregated from the test split for [PITH_FULL_IMAGE:figures/full_fig_p011_4.png] view at source ↗

**Figure 5.** Figure 5: Qualitative SR comparison on examples degraded at scales [PITH_FULL_IMAGE:figures/full_fig_p012_5.png] view at source ↗

**Figure 6.** Figure 6: Qualitative comparison for day-to-night and night-to-day translation using Cycle [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗

read the original abstract

Recent advances in deep learning for remote sensing rely heavily on large annotated datasets, yet acquiring high-quality ground truth for geometric, radiometric, and multi-domain tasks remains costly and often infeasible. In particular, the lack of accurate depth annotations, controlled illumination variations, and multi-scale paired imagery limits progress in monocular depth estimation, domain adaptation, and super-resolution for aerial scenes. We present SyMTRS, a large-scale synthetic dataset generated using a high-fidelity urban simulation pipeline. The dataset provides high-resolution RGB aerial imagery (2048 x 2048), pixel-perfect depth maps, night-time counterparts for domain adaptation, and aligned low-resolution variants for super-resolution at x2, x4, and x8 scales. Unlike existing remote sensing datasets that focus on a single task or modality, SyMTRS is designed as a unified multi-task benchmark enabling joint research in geometric understanding, cross-domain robustness, and resolution enhancement. We describe the dataset generation process, its statistical properties, and its positioning relative to existing benchmarks. SyMTRS aims to bridge critical gaps in remote sensing research by enabling controlled experiments with perfect geometric ground truth and consistent multi-domain supervision. The results obtained in this work can be reproduced from this Github repository: https://github.com/safouaneelg/SyMTRS.

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.

Desk Editor's Note private letter to a colleague

SyMTRS gives researchers a single synthetic package with aligned RGB, depth, night-time pairs, and multi-scale low-res targets for aerial scenes, but the paper offers no measurements showing the simulator matches real data or that models trained on it transfer.

read the letter

The core contribution is a new synthetic dataset that bundles high-res 2048x2048 aerial RGB, pixel-perfect depth, night-time versions for domain adaptation, and downsampled pairs for 2x/4x/8x super-resolution, all generated from one urban simulation pipeline. They release the data and code on GitHub, describe the generation steps, note some statistical properties, and compare it to prior single-task remote-sensing datasets. That combination in one consistent resource is the actual novelty; most existing sets handle depth or domain shift or resolution separately, so this lowers the friction for multi-task experiments where real ground truth is hard to get at scale. The release itself is useful because dataset papers live or die on whether people can actually download and use the data without extra hurdles. The positioning against other benchmarks is straightforward and helps readers see where it fits. The main weakness is the lack of any quantitative validation. The abstract and description claim high fidelity and realistic domain shifts, but there are no numbers on image statistics versus real aerial photos, no depth error metrics against LiDAR, and no transfer experiments training on SyMTRS then testing on real imagery. Without those, the utility claim stays untested. The paper is mostly descriptive rather than evaluative, which is common for dataset releases but leaves the central assumption about simulator realism unexamined. This is for remote-sensing groups that want controlled multi-task data to prototype models before moving to real captures, or for anyone studying synthetic-to-real gaps in depth estimation and super-resolution. A reader who needs perfect labels across tasks will find it practical to start with. It is worth sending to peer review because the data artifact is new and released, and referees can push for the missing validation experiments that would make the benchmark more credible. Desk rejection would be too quick given the practical gap it targets.

Referee Report

3 major / 2 minor

Summary. The manuscript introduces SyMTRS, a large-scale synthetic multi-task dataset for aerial imagery generated via a high-fidelity urban simulation pipeline. It supplies 2048x2048 RGB images paired with pixel-perfect depth maps, night-time domain-shift counterparts, and aligned low-resolution variants at x2/x4/x8 scales for super-resolution, positioning the resource as a unified benchmark for joint work on monocular depth estimation, cross-domain adaptation, and resolution enhancement in remote sensing.

Significance. If the simulation produces imagery whose statistical properties and domain shifts are representative of real aerial data, the dataset would enable controlled multi-task experiments with perfect geometric ground truth that real remote-sensing collections rarely provide. The accompanying GitHub repository for reproduction is a clear strength that supports benchmark adoption.

major comments (3)

[Abstract and dataset-generation description] Abstract and dataset-generation description: the repeated claim of 'pixel-perfect depth maps' and 'high-fidelity' urban simulation is not accompanied by any quantitative validation (e.g., depth-error histograms against known simulation parameters or comparison to real LiDAR statistics), leaving the central realism assumption untested.
[Statistical-properties and positioning section] Statistical-properties and positioning section: no tables or figures report concrete similarity metrics (FID, depth-distribution KL divergence, or day/night radiometric shift measures) between SyMTRS and existing real or synthetic aerial benchmarks, undermining the claim that the dataset bridges gaps for domain-adaptation and multi-scale research.
[Overall contribution] Overall contribution: the manuscript contains no baseline experiments (e.g., depth-estimation or SR transfer results from SyMTRS to a real test set), so the assertion that the resource 'enables joint research' rests solely on the pipeline description rather than demonstrated utility.

minor comments (2)

Verify that all figure captions explicitly state image dimensions, scale factors, and whether night-time pairs are aligned at the pixel level.
Add a short table summarizing key simulation parameters (camera intrinsics, lighting model, urban asset density) to improve reproducibility beyond the GitHub link.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback. We address each major comment point by point below, acknowledging where the manuscript can be strengthened through revision while providing honest clarification on the dataset's design and contribution.

read point-by-point responses

Referee: [Abstract and dataset-generation description] Abstract and dataset-generation description: the repeated claim of 'pixel-perfect depth maps' and 'high-fidelity' urban simulation is not accompanied by any quantitative validation (e.g., depth-error histograms against known simulation parameters or comparison to real LiDAR statistics), leaving the central realism assumption untested.

Authors: We clarify that 'pixel-perfect' depth is obtained directly from the simulation engine's 3D geometry, yielding exact per-pixel values by construction without the reconstruction or sensor errors present in real LiDAR. We agree, however, that quantitative validation would strengthen the realism claims. In the revised manuscript we will add depth-error histograms derived from known simulation parameters together with depth-distribution comparisons against publicly available real aerial LiDAR statistics. revision: yes
Referee: [Statistical-properties and positioning section] Statistical-properties and positioning section: no tables or figures report concrete similarity metrics (FID, depth-distribution KL divergence, or day/night radiometric shift measures) between SyMTRS and existing real or synthetic aerial benchmarks, undermining the claim that the dataset bridges gaps for domain-adaptation and multi-scale research.

Authors: The manuscript contains a dedicated section describing statistical properties and relative positioning, yet we acknowledge the absence of the specific quantitative metrics mentioned. We will incorporate FID scores, depth-distribution KL divergences, and day/night radiometric shift measures, along with the corresponding tables and figures, in the revised version to provide stronger empirical support for the dataset's utility in domain-adaptation and multi-scale tasks. revision: yes
Referee: [Overall contribution] Overall contribution: the manuscript contains no baseline experiments (e.g., depth-estimation or SR transfer results from SyMTRS to a real test set), so the assertion that the resource 'enables joint research' rests solely on the pipeline description rather than demonstrated utility.

Authors: As a dataset-introduction paper, the core contribution lies in the generation pipeline, perfect ground-truth annotations, and public release that together enable controlled multi-task experiments. We nevertheless agree that preliminary baseline results would better illustrate practical utility. In the revision we will add baseline experiments for monocular depth estimation and super-resolution, including limited transfer results from SyMTRS to a real aerial test set. revision: yes

Circularity Check

0 steps flagged

No circularity: dataset artifact with no derivations or fitted predictions

full rationale

The paper introduces SyMTRS as a synthetic multi-task dataset generated from an urban simulation pipeline. No equations, parameter fitting, predictions, or derivation chains appear in the abstract or described content. The core contribution is the dataset artifact (RGB, depth, night-time, and multi-scale variants) rather than any computed result that could reduce to its own inputs by construction. Self-citations or uniqueness claims are absent from the provided text. This matches the default expectation for non-circular papers and the reader's 0.0 assessment.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

No mathematical derivations, fitted parameters, or postulated entities are introduced; the paper's contribution is the creation and release of a synthetic dataset generated via an existing simulation pipeline.

pith-pipeline@v0.9.0 · 5552 in / 1102 out tokens · 24439 ms · 2026-05-09T22:43:49.843654+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

49 extracted references · 18 canonical work pages

[1]

Virtual kitti 2, in: CVPR

Cabon, Y., Murray, N., Humenberger, M., 2020. Virtual kitti 2, in: CVPR

2020
[2]

M3vir: Multi-modal multi-task multi-view immersive rendering dataset, in: CVPR

Chen, X., Zhang, Y., Xu, J., et al., 2024a. M3vir: Multi-modal multi-task multi-view immersive rendering dataset, in: CVPR
[3]

M3vir: A multi-modal multi-task multi-view immersive rendering dataset, in: CVPR

Chen, X., Zhang, Y., Xu, J., et al., 2024b. M3vir: A multi-modal multi-task multi-view immersive rendering dataset, in: CVPR
[4]

Remote sensing image scene classification: Benchmark and state of the art

Cheng, G., Han, J., Lu, X., 2017. Remote sensing image scene classification: Benchmark and state of the art. Proceedings of the IEEE . 13

2017
[5]

Image super- resolution with deep variational autoencoders, in: Karlinsky, L., Michaeli, T., Nishino, K

Chira, D., Haralampiev, I., Winther, O., Dittadi, A., Liévin, V ., 2023. Image super- resolution with deep variational autoencoders, in: Karlinsky, L., Michaeli, T., Nishino, K. (Eds.), Computer Vision – ECCV 2022 Workshops, Springer Nature Switzerland, Cham. pp. 395–411

2023
[6]

Functional map of the world

Christie, G., Fendley, N., Wilson, J., Mukherjee, R., 2018. Functional map of the world. CVPR Workshops

2018
[7]

The cityscapes dataset for semantic urban scene understand- ing, in: CVPR

Cordts, M., Omran, M., Ramos, S., Rehfeld, T., Enzweiler, M., Benenson, R., Franke, U., Roth, S., Schiele, B., 2016. The cityscapes dataset for semantic urban scene understand- ing, in: CVPR

2016
[8]

Deepglobe 2018: A challenge to parse the earth through satellite images

Demir, I., Koperski, K., Lindenbaum, D., Pang, G., Huang, J., Basu, S., Hughes, F., Tuia, D., Raska, R., 2018. Deepglobe 2018: A challenge to parse the earth through satellite images. CVPR Workshops

2018
[9]

Rareplanes: Synthetic data to improve aircraft detection in satellite imagery

DIU, Works, I.Q.T.C., 2020. Rareplanes: Synthetic data to improve aircraft detection in satellite imagery. https://www.cosmiqworks.org/rareplanes/

2020
[10]

Image super-resolution using deep convolu- tional networks

Dong, C., Loy, C.C., He, K., Tang, X., 2016. Image super-resolution using deep convolu- tional networks. IEEE Transactions on Pattern Analysis and Machine Intelligence 38, 295–307. doi:10.1109/TPAMI.2015.2439281

work page doi:10.1109/tpami.2015.2439281 2016
[11]

Rrsgan: Reference-based super-resolution for remote sensing image

Dong, R., Lixian, Z., Fu, H., 2021. Rrsgan: Reference-based super-resolution for remote sensing image. IEEE Transactions on Geoscience and Remote Sensing PP , 1–17. doi:10.1109/TGRS.2020.3046045

work page doi:10.1109/tgrs.2020.3046045 2021
[12]

The pascal visual object classes (voc) challenge

Everingham, M., Van Gool, L., Williams, C.K., Winn, J., Zisserman, A., 2010. The pascal visual object classes (voc) challenge. International journal of computer vision

2010
[13]

Mid-air: A multi-modal dataset for extremely low altitude drone flights, in: IROS

Fonder, M., Courbon, J., et al., 2019. Mid-air: A multi-modal dataset for extremely low altitude drone flights, in: IROS

2019
[14]

Vision meets robotics: The kitti dataset, in: The International Journal of Robotics Research

Geiger, A., Lenz, P ., Urtasun, R., 2013. Vision meets robotics: The kitti dataset, in: The International Journal of Robotics Research

2013
[15]

Eurosat: A novel dataset and deep learning benchmark for land use and land cover classification

Helber, P ., Bischke, B., Dengel, A., Borth, D., 2019. Eurosat: A novel dataset and deep learning benchmark for land use and land cover classification. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing

2019
[16]

Image-to-Image Translation with Conditional Adversarial Networks , journal =

Isola, P ., Zhu, J.Y., Zhou, T., Efros, A.A., 2017. Image-to-image translation with conditional adversarial networks, in: Proceedings of the IEEE Con- ference on Computer Vision and Pattern Recognition (CVPR). URL: https: //openaccess.thecvf.com/content_cvpr_2017/papers/Isola_Image-To-Image_ Translation_With_CVPR_2017_paper.pdf. arXiv:1611.07004

work page arXiv 2017
[17]

Isprs potsdam dataset

ISPRS, 2018. Isprs potsdam dataset. https://www2.isprs.org/commissions/comm2/ wg4/potsdam-2d-semantic-labeling/

2018
[18]

Imagenet classification with deep convolutional neural networks

Krizhevsky, A., Sutskever, I., Hinton, G.E., 2012. Imagenet classification with deep convolutional neural networks. Communications of the ACM

2012
[19]

xview: Objects in context in overhead imagery doi: 10.48550/arXiv.1802

Lam, D., Kuzma, R., McGee, K., Dooley, S., Laielli, M., Klaric, M., Bulatov, Y., McCord, B., 2018. xview: Objects in context in overhead imagery doi: 10.48550/arXiv.1802. 07856

work page doi:10.48550/arxiv.1802 2018
[20]

Photo-realistic single image super-resolution using a generative adversarial network, pp

Ledig, C., Theis, L., Huszar, F., Caballero, J., Cunningham, A., Acosta, A., Aitken, A., Tejani, A., Totz, J., Wang, Z., Shi, W., 2017. Photo-realistic single image super-resolution using a generative adversarial network, pp. 105–114. doi:10.1109/CVPR.2017.19

work page doi:10.1109/cvpr.2017.19 2017
[21]

2023 , url =

Li, Y., Jiang, L., Xu, L., Xiangli, Y., Wang, Z., Lin, D., Dai, B., 2023. Matrixcity: A large-scale city dataset for city-scale neural rendering and beyond, pp. 3182–3192. doi:10.1109/ICCV51070.2023.00297

work page doi:10.1109/iccv51070.2023.00297 2023
[22]

In2021 IEEE/CVF International Conference on Computer Vision Workshops (ICCVW)

Liang, J., Cao, J., Sun, G., Zhang, K., Van Gool, L., Timofte, R., 2021. Swinir: Image restoration using swin transformer, in: 2021 IEEE/CVF International Conference on Computer Vision Workshops (ICCVW), pp. 1833–1844. doi:10.1109/ICCVW54120.2021. 00210. 14

work page doi:10.1109/iccvw54120.2021 2021
[23]

A comparative study of deep learning models for image super-resolution, in: Jiang, X

Lim, J.Y., Chiew, Y.S., Phan, R.C.W., Wang, X., 2024. A comparative study of deep learning models for image super-resolution, in: Jiang, X. (Ed.), Asia Conference on Elec- tronic Technology (ACET 2024), International Society for Optics and Photonics. SPIE. p. 1321105. URL:https://doi.org/10.1117/12.3032724, doi:10.1117/12.3032724

work page doi:10.1117/12.3032724 2024
[24]

Microsoft coco: Common objects in context

Lin, T.Y., Maire, M., Belongie, S., et al., 2014. Microsoft coco: Common objects in context. ECCV

2014
[25]

Usegeo - a uav-based multi-sensor dataset for geospatial research

Nex, F., Stathopoulou, E., Remondino, F., Yang, M., Madhuanand, L., Yogender, Y., Alsadik, B., Weinmann, M., Jutzi, B., Qin, R., 2024. Usegeo - a uav-based multi-sensor dataset for geospatial research. ISPRS Open Journal of Photogrammetry and Remote Sensing 13, 100070. URL: https://www.sciencedirect.com/science/article/pii/ S2667393224000140, doi:https://...

work page doi:10.1016/j.ophoto.2024.100070 2024
[26]

A comparative analysis of srgan models

Nikroo, F.R., Deshmukh, A., Sharma, A., Tam, A., Kumar, K., Norris, C., Dangi, A., 2023. A comparative analysis of srgan models. URL: https://arxiv.org/abs/2307.09456, arXiv:2307.09456

work page arXiv 2023
[27]

The synthia dataset: A large collection of synthetic images for semantic segmentation of urban scenes

Ros, G., Sellart, L., Materzynska, J., Vazquez, D., Lopez, A.M., 2016. The synthia dataset: A large collection of synthetic images for semantic segmentation of urban scenes. CVPR

2016
[28]

Airsim: High-fidelity visual and physical simulation for autonomous vehicles

Shah, S., Dey, D., et al., 2017. Airsim: High-fidelity visual and physical simulation for autonomous vehicles. Field and service robotics

2017
[29]

Indoor segmentation and support inference from rgbd images, in: ECCV

Silberman, N., Hoiem, D., Kohli, P ., Fergus, R., 2012. Indoor segmentation and support inference from rgbd images, in: ECCV

2012
[31]

Synrs3d: A synthetic multi-task benchmark for remote sensing 3d understanding

Song, Y., Zhang, W., Wang, T., et al., 2024b. Synrs3d: A synthetic multi-task benchmark for remote sensing 3d understanding. arXiv preprint arXiv:2409.05142

work page arXiv
[32]

Soni, J., Gurappa, S., Upadhyay, H., 2024. A comparative study of deep learning models for image super-resolution across various magnification levels, in: 2024 IEEE International Conference on Future Machine Learning and Data Science (FMLDS), pp. 395–400. doi:10.1109/FMLDS63805.2024.00076

work page doi:10.1109/fmlds63805.2024.00076 2024
[33]

Bigearthnet: A large-scale benchmark archive for remote sensing image understanding

Sumbul, G., Charfuelan, M., Demir, B., Markl, V ., 2019. Bigearthnet: A large-scale benchmark archive for remote sensing image understanding. IGARSS

2019
[34]

Shift: A synthetic driving dataset for domain adaptation and generalization, in: CVPR

Sun, K., Liu, Z., et al., 2023. Shift: A synthetic driving dataset for domain adaptation and generalization, in: CVPR

2023
[35]

Deep satellite video super-resolution via global registration and local alignment

Wang, K., Wu, F., Luo, X., et al., 2022. Deep satellite video super-resolution via global registration and local alignment. CVPR

2022
[36]

Tartanair: A dataset to push the limits of visual slam

Wang, Y., Liu, Y., et al., 2020. Tartanair: A dataset to push the limits of visual slam. arXiv preprint arXiv:2003.14338

work page arXiv 2020
[37]

Loveda: A remote sensing land cover dataset for domain adaptive semantic segmentation, in: NeurIPS

Wang, Y., Mao, J., et al., 2021a. Loveda: A remote sensing land cover dataset for domain adaptive semantic segmentation, in: NeurIPS
[38]

Loveda: A remote sensing land cover dataset for domain adaptive semantic segmentation, in: NeurIPS

Wang, Y., Mao, J., Yu, X., Jin, Y., Li, X., Sun, L., 2021b. Loveda: A remote sensing land cover dataset for domain adaptive semantic segmentation, in: NeurIPS
[39]

InProceedings of the IEEE/CVF conference on computer vision and pattern recognition

Wang, Z., Bovik, A., Sheikh, H., Simoncelli, E., 2004. Image quality assessment: from error visibility to structural similarity. IEEE Transactions on Image Processing 13, 600–612. doi:10.1109/TIP.2003.819861

work page doi:10.1109/tip.2003.819861 2004
[41]

Samrs: Supervised pretraining for remote sensing foundation models

Wang, Z., Liu, Q., Yu, L., et al., 2024b. Samrs: Supervised pretraining for remote sensing foundation models. arXiv preprint arXiv:2506.23801

work page arXiv
[42]

Oli2msi: A multi-sensor super-resolution dataset for remote sensing

Wei, Y., Zhang, H., Peng, X., Xu, Y., Wang, Z., Li, Y., 2021. Oli2msi: A multi-sensor super-resolution dataset for remote sensing. IGARSS . 15

2021
[43]

Dota: A large-scale dataset for object detection in aerial images

Xia, G.S., Bai, X., Ding, J., Zhu, Z., Belongie, S., Luo, J., 2018. Dota: A large-scale dataset for object detection in aerial images. CVPR

2018
[44]

Aid: A benchmark dataset for performance evaluation of aerial scene classification

Xia, G.S., Hu, J., Hu, F., Shi, B., Bai, X., Zhong, Y., Zhang, L., Lu, X., 2017. Aid: A benchmark dataset for performance evaluation of aerial scene classification. IEEE Transactions on Geoscience and Remote Sensing

2017
[45]

Wilduav: Real uav flight data for aerial scene understanding

Xie, J., et al., 2023. Wilduav: Real uav flight data for aerial scene understanding. Remote Sensing

2023
[46]

Bag-of-visual-words and spatial extensions for land-use classification

Yang, Y., Newsam, S., 2010. Bag-of-visual-words and spatial extensions for land-use classification. ACM SIGSPATIAL

2010
[47]

Bdd100k: A diverse driving dataset for heterogeneous multitask learning, in: CVPR

Yu, F., Chen, H., Wang, X., Xian, W., Chen, Y., Liu, F., Madhavan, V ., Darrell, T., 2020. Bdd100k: A diverse driving dataset for heterogeneous multitask learning, in: CVPR

2020
[48]

A comparative study of deep learn- ing methods for super-resolution of npp-viirs nighttime light images

Zhang, C., Mao, Z., Nie, J., Lai, Y., Deng, L., 2025. A comparative study of deep learn- ing methods for super-resolution of npp-viirs nighttime light images. International Journal of Applied Earth Observation and Geoinformation 145, 104995. URL: https:// www.sciencedirect.com/science/article/pii/S1569843225006429, doi: https:// doi.org/10.1016/j.jag.2025.104995

work page doi:10.1016/j.jag.2025.104995 2025
[49]

Places: A 10 million image database for scene recognition, in: IEEE Transactions on Pattern Analysis and Machine Intelligence

Zhou, B., Lapedriza, A., Khosla, A., Oliva, A., Torralba, A., 2017. Places: A 10 million image database for scene recognition, in: IEEE Transactions on Pattern Analysis and Machine Intelligence

2017
[50]

Sen2naip: A real-world benchmark for cross-sensor super-resolution

Zhou, T., Wang, Y., Duan, K., Xu, Q., Tu, Z., 2023. Sen2naip: A real-world benchmark for cross-sensor super-resolution. arXiv preprint arXiv:2311.09756

work page arXiv 2023
[51]

Unpaired Image- to-Image Translation using Cycle-Consistent Adversarial Networks,

Zhu, J.Y., Park, T., Isola, P ., Efros, A.A., 2017. Unpaired image-to-image translation using cycle-consistent adversarial networks, in: Proceedings of the IEEE International Conference on Computer Vision (ICCV). URL: https://openaccess.thecvf.com/content_ICCV_2017/papers/Zhu_Unpaired_ Image-To-Image_Translation_ICCV_2017_paper.pdf. arXiv:1703.10593. 16

work page arXiv 2017