Pith. sign in

REVIEW 3 major objections 6 minor 33 references

A lightweight hybrid GAN turns game-engine renders into photoreal images at real-time speed without wrecking scene content.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-14 23:28 UTC pith:Q6OY6CXX

load-bearing objection Solid real-time engineering paper: compact U-Net + hybrid real-patch adversarial batches give a clear ~6× 1080p FPS win and modest quality/semantic gains over FastCUT/REGEN and paired-only training, with public code. the 3 major comments →

arxiv 2603.10604 v3 pith:Q6OY6CXX submitted 2026-03-11 cs.CV

HyPER-GAN: Hybrid Patch-Based Image-to-Image Translation for Real-Time Photorealism Enhancement in Game Engines

Stefanos Pasios , Nikos Nikolaidis This is my paper
classification cs.CV
keywords photorealism enhancementimage-to-image translationreal-time GANsim-to-realU-Net generatorhybrid patch trainingGTA-Vsynthetic data
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Game engines and simulators produce large labeled image sets that look synthetic, so computer-vision models trained on them often fail in the real world. Heavy generative models can close that appearance gap, but they are too slow for real-time use and frequently invent objects or textures that were never in the original scene. HyPER-GAN answers both problems with a compact U-Net generator that runs above 30 frames per second at 1080p on a mid-range GPU. It is trained on synthetic–photorealism pairs plus carefully matched patches drawn from real photographs, so the network learns real-world look while being discouraged from copying the teacher’s artifacts. The result is faster, more photorealistic, and more content-preserving translation than previous lightweight paired methods, and the same model transfers across different game engines.

Core claim

HyPER-GAN shows that a lightweight U-Net generator, trained with a hybrid adversarial objective that mixes paired photorealism-enhanced images and FAISS-matched real-world patches, can raise the visual realism of synthetic game-engine frames to real-time rates (≈34 FPS at 1080p) while preserving semantic content better than existing lightweight paired translators and better than the same architecture trained only on the paired images.

What carries the argument

Hybrid patch training: four non-overlapping 196×196 patches are taken from each generated image; their nearest real-world neighbors (VGG-16 features, FAISS L2) are mixed with the corresponding teacher patches so the PatchGAN discriminator must accept both domains, forcing the generator away from teacher artifacts.

Load-bearing premise

The method assumes that nearest-neighbor real patches found by VGG-16 features are clean enough teachers that the compact network will learn real appearance without simply absorbing the artifacts already present in the photorealism-enhanced pairs.

What would settle it

Train the identical U-Net solely on the paired synthetic–EPE images (HyPER-GAN-EO) versus the hybrid objective; if hybrid training does not measurably lower KID to the real datasets and raise Mask2Former mIoU on the same test split, the hybrid claim fails.

Watch this falsifier — get emailed when new claim-graph text bears on it.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The paper proposes HyPER-GAN, a lightweight U-Net-style paired image-to-image translator for real-time photorealism enhancement of synthetic game/simulator images. It is trained on synthetic–EPE proxy pairs and a hybrid PatchGAN objective that mixes EPE target patches with FAISS-matched real-world patches (VGG-16 block-4 features, 196×196 non-overlapping patches). On GTA-V/PFD→CS/MV, the method reports ~6× higher FPS at 1080p than FastCUT/REGEN (Table 1: 33.7 vs 5.5 FPS on RTX 4070 Super), lower KID and higher Mask2Former mIoU than those baselines and than a paired-only ablation (HyPER-GAN-EO; Table 2), qualitative reduction of EPE-style artifacts (Figs. 3–4), better KID/mAP than REGEN and COSMOS Transfer1 on PFB (Table 3), and competitive cross-engine transfer to CARLA-UE5 (Table 4). Code and models are released.

Significance. If the results hold, the work is a useful systems contribution for synthetic-data pipelines and game-engine post-processing: a compact feed-forward generator that reaches real-time 1080p on mid-range GPUs while improving KID and semantic metrics over recent lightweight paired baselines, without G-buffers at inference. Strengths include public code/models, multi-metric evaluation (latency/VRAM, KID, Mask2Former mIoU, YOLO mAP), an explicit paired-only ablation, and cross-engine/diffusion comparisons. The hybrid real-patch idea is a practical way to reuse unpaired real data when only teacher-generated pairs exist. Significance is incremental rather than foundational—it depends on EPE-style teachers and modest hybrid gains—but the speed–quality tradeoff is relevant to sim2real and real-time rendering communities.

major comments (3)
  1. §2.2–2.3 and Eq. (5)–(6): The central content-protection claim is that FAISS L2 nearest neighbors of VGG-16 block-4 features on 196×196 patches, mixed into the PatchGAN real set with EPE targets, prevent the generator from inheriting teacher artifacts while the L1 term still pulls toward EPE. Match quality is never measured (class agreement, retrieval rank, fraction of true content neighbors vs texture-only matches), nor is residual EPE artifact rate in the paired targets quantified. Table 2 hybrid gains over EO are modest (CS KID 3.41 vs 4.06; mIoU 48.79% vs 47.04%), and Fig. 3 is qualitative. Without a match-quality or artifact-inheritance analysis, the mechanism that is said to deliver superior semantic robustness remains under-supported; the speed claim can stand, but quality superiority over lightweight paired methods needs stronger evidence that hybrid supervision is doing the clai
  2. §3.1 and Table 2: Semantic robustness is measured by applying Mask2Former models pretrained on CS/MV to generated images and comparing mIoU to PFD ground truth (19 CS classes). This is a reasonable proxy, but it does not isolate content preservation from domain shift of the segmenter itself, and incompatibilities with PFD labels are handled by collapsing classes to background without reporting per-class IoU or failure modes. Given that hybrid gains over EO and REGEN are small in absolute mIoU, the paper should either report per-class breakdowns / qualitative failure cases or add a content-consistency metric less dependent on a real-domain segmenter (e.g., synthetic-label agreement under a fixed synthetic-trained segmenter, or LPIPS/feature distance to the synthetic input under a content mask). Otherwise the claim of “without significantly compromising semantic consistency” rests on a sin
  3. Appendix B / Table 4 and abstract claim of improved cross-engine robustness: HyPER-GAN is trained on PFD (GTA-V) while REGEN/EPE baselines for CARLA-UE5 were trained on CARLA-UE4; training set sizes also differ (9,549 vs 15,011). The abstract states that hybrid training “increases the robustness in cross-engine evaluations” relative to paired-only training, but Table 4 does not report HyPER-GAN-EO on CARLA-UE5, so that specific claim is not evidenced. Either add the EO cross-engine numbers or narrow the abstract/conclusion wording to what is actually measured (competitive transfer of the hybrid model vs REGEN/EPE under mismatched training domains).
minor comments (6)
  1. Abstract vs body: the abstract claims temporal consistency is maintained, but the main experimental section does not report a temporal metric (e.g., warping error or flicker on PFB sequences); only Appendix A uses video inputs for COSMOS. Add a short quantitative temporal check or qualify the claim.
  2. §2.1 preprocessing: images are resized to 512×512 for training while evaluation is at native/1080p; briefly state whether the generator is fully convolutional and how resolution is handled at inference to avoid ambiguity.
  3. Eq. (3) decoder: skip connections are written as concatenations with e2/e1, but the first decoder stage has no skip from e3; a one-line architecture diagram or channel table would make the U-Net layout easier to reproduce.
  4. Table 1 reports mean±std over 100 images without specifying warm-up, batching, or precision (FP32/FP16); a short note would improve reproducibility of the 6× FPS claim.
  5. Author-block contact lines appear swapped (pstefanos vs nnik emails); fix for the camera-ready version.
  6. Related-work framing of REGEN and EPE is clear, but a short note on why other real-time style-transfer / G-buffer methods (beyond the two paired baselines) are out of scope would help readers place the contribution.

Circularity Check

0 steps flagged

No circular derivation: empirical Im2Im method with external KID/mIoU evaluation; self-citations are baselines/data, not load-bearing premises.

full rationale

HyPER-GAN is an engineering/empirical CV paper, not a first-principles derivation. The generator (lightweight U-Net, Eqs. 1–3), PatchGAN discriminator (Eq. 4), LSGAN+L1 objectives (Eq. 6), and FAISS/VGG-16 patch matching (Eq. 5, following EPE) are standard design choices whose success is measured by external benchmarks: KID against real CS/MV, mIoU from independently pretrained Mask2Former, FPS/VRAM on fixed hardware, and cross-engine transfer to CARLA-UE5. Training on EPE proxy pairs is teacher-dependent distillation, not circularity—the paper does not claim to predict EPE outputs or redefine photorealism as the training loss; it reports lower KID and higher mIoU than FastCUT/REGEN/EO on held-out real-domain metrics, and qualitatively fewer EPE artifacts. Self-citations (REGEN as baseline and conceptual predecessor; CARLA2Real/CARLA-UE5 for evaluation assets) supply comparison methods and datasets, not uniqueness theorems or fitted parameters renamed as predictions. No step reduces a claimed result to its inputs by construction. Score 0 with empty steps is the correct finding.

Axiom & Free-Parameter Ledger

5 free parameters · 4 axioms · 1 invented entities

This is an empirical GAN paper. Load-bearing content is architectural and training choices plus the assumption that EPE pairs and VGG-matched real patches are valid supervision. Free parameters are standard training hyperparameters; axioms are domain assumptions of adversarial Im2Im and feature-space patch similarity; no new physical entities are postulated.

free parameters (5)
  • L1 reconstruction weight λ = 10
    Set to 10 in Eq. 6 without a sensitivity study; scales content preservation vs adversarial realism.
  • Generator/discriminator learning rate and Adam betas = 2e-4; (0.5, 0.999)
    2e-4 with betas (0.5, 0.999); standard but chosen by hand for this setup.
  • Training epochs / batch size = 20 epochs, bs=1 (HyPER-GAN)
    20 epochs, batch size 1 for HyPER-GAN; REGEN baseline trained 9 epochs as in prior work—training budget is a free experimental choice affecting comparisons.
  • Patch size and count for matching = 4 × 196×196
    Four non-overlapping 196×196 patches after 512×512 resize; defines the hybrid real-world supervision granularity.
  • U-Net channel widths and residual depth = 64-128-256; 4 ResBlocks
    Encoder channels 64/128/256 and four residual blocks are design choices that trade capacity for real-time speed.
axioms (4)
  • domain assumption LSGAN adversarial training plus L1 to a teacher-enhanced target yields a generator that improves photorealism while preserving structure.
    Standard paired Im2Im assumption invoked in Sec. 2.3 / Eq. 6; not proved, only empirically tested.
  • domain assumption VGG-16 block-4 conv features with L2 nearest neighbors identify semantically corresponding real patches useful as additional real adversarial targets.
    Core of hybrid strategy (Sec. 2.2–2.3, Eq. 5); borrowed from EPE-style matching and treated as valid supervision.
  • domain assumption KID to CS/MV and mIoU/mAP of real-pretrained segmenters/detectors are adequate proxies for visual realism and semantic consistency of enhanced frames.
    Evaluation design in Sec. 3.1 and appendices; common in the subfield but not equivalent to human or task-end-to-end validation.
  • ad hoc to paper EPE outputs are usable paired proxy targets for the real domain despite known artifacts.
    Method builds on REGEN’s reformulation; hybrid real patches are introduced specifically to mitigate this axiom’s failure modes.
invented entities (1)
  • HyPER-GAN hybrid training objective (paired EPE patches + matched real patches in PatchGAN batches) no independent evidence
    purpose: Provide real-time photorealism enhancement that avoids learning teacher artifacts while remaining lightweight.
    Not a physical entity; a procedural construct. Independent evidence is only the paper’s own ablations and metrics, not an external falsifiable prediction outside this setup.

pith-pipeline@v1.1.0-grok45 · 16332 in / 3503 out tokens · 35679 ms · 2026-07-14T23:28:46.936496+00:00 · methodology

0 comments
read the original abstract

Generative models are increasingly used in video game engines to enhance the photorealism of rendered images for visual synthetic data generation and simulation applications. However, they often introduce artifacts that alter the content of the original rendered scenes and require high computational resources, which limit their utilization for the photorealism enhancement of training and evaluation data, as well as their integration in the rendering pipelines of game engines. In this paper, we propose Hybrid Patch Enhanced Realism Generative Adversarial Network (HyPER-GAN), a hybrid image-to-image translation framework that is based on a lightweight U-Net-style generator capable of performing real-time inference. The framework is trained using paired rendered and photorealism-enhanced images, complemented by a novel hybrid training strategy that incorporates matched patches from unpaired real-world images to improve content preservation and further enhance the visual realism that can be achieved by the lightweight generator. Experimental results demonstrate that HyPER-GAN achieves a 6x increase in frames per second at 1080p in comparison with state-of-the-art lightweight paired image-to-image translation methods, while also increasing, in both within- and cross-engine evaluations, the photorealism of the rendered images without significantly compromising semantic consistency. Moreover, it is illustrated that HyPER-GAN maintains temporal consistency and that the proposed hybrid training strategy improves content preservation and visual realism in within-engine and increases the robustness in cross-engine evaluations compared to training the framework solely with paired rendered and photorealism-enhanced images. Code and pretrained models are publicly available at: https://github.com/stefanos50/HyPER-GAN

Figures

Figures reproduced from arXiv: 2603.10604 by Nikos Nikolaidis, Stefanos Pasios.

Figure 1
Figure 1. Figure 1: Overview of the HyPER-GAN method, which includes four phases: a) datasets and preprocessing, b) real-world dataset indexing, c) training, and d) photorealism enhancement. potentially produced during the unpaired Im2Im translation phase [26]. In this work, we build upon the paired Im2Im concept introduced by REGEN and propose Hybrid Patch Enhanced Realism Generative Adversarial Network (HyPER-GAN), a lightw… view at source ↗
Figure 2
Figure 2. Figure 2: Examples of matched patches between the generated (top) and the real-world (bottom). 𝐺 from learning artifacts introduced by the robust unpaired Im2Im translation network, as it must fool 𝐷 not only on the photorealism-enhanced target domain but also on the real-world one, where such artifacts are absent. For com￾parison, we also consider a simplified variation, HyPER￾GAN-EO, in which 𝐺 is trained exclusiv… view at source ↗
Figure 4
Figure 4. Figure 4: Translation results of PFD (GTA-V) towards the real-world datasets CS and MV produced by a) FastCUT, b) REGEN, and c) HyPER-GAN. towards CS are evaluated against CS) using the Kernel In￾ception Distance (KID) [3] metric. To investigate the seman￾tic preservation capabilities of the proposed HyPER-GAN, the HyPER-GAN-EO variation, and the baseline methods, Mask2Former [4] semantic segmentation models pretrai… view at source ↗
Figure 5
Figure 5. Figure 5: Comparison of images generated by (b) COSMOS Transfer1 and (c) HyPER-GAN, given as input an image (a) from the PFB dataset. [2] Batziou, E., Ioannidis, K., Patras, I., Vrochidis, S., Kompatsiaris, I., 2026. Hdd-unet: A unet-based architecture for low-light image enhancement. Image and Vision Computing 167, 105889. doi:https: //doi.org/10.1016/j.imavis.2025.105889. [3] Bińkowski, M., Sutherland, D.J., Arbel… view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

33 extracted references · 6 canonical work pages

  1. [1]

    Cosmos-transfer1: Conditional world generation with adaptive multimodal control.arXiv:2503.14492

    Alhaija, H.A., et al., 2025. Cosmos-transfer1: Conditional world generation with adaptive multimodal control.arXiv:2503.14492. S. Pasios et al.:Preprint submitted to ElsevierPage 6 of 7 HyPER-GAN: Real-Time Photorealism Enhancement Figure 5:Comparison of images generated by (b) COSMOS Transfer1 and (c) HyPER-GAN, given as input an image (a) from the PFB dataset

    Pith/arXiv arXiv 2025

  2. [2]

    Hdd-unet: A unet-based architecture for low-light image enhancement

    Batziou, E., Ioannidis, K., Patras, I., Vrochidis, S., Kompatsiaris, I., 2026. Hdd-unet: A unet-based architecture for low-light image enhancement. ImageandVisionComputing167,105889. doi:https: //doi.org/10.1016/j.imavis.2025.105889

    work page doi:10.1016/j.imavis.2025.105889 2026

  3. [3]

    De- mystifying mmd gans.arXiv:1801.01401

    Bińkowski, M., Sutherland, D.J., Arbel, M., Gretton, A., 2021. De- mystifying mmd gans.arXiv:1801.01401

    Pith/arXiv arXiv 2021

  4. [4]

    Masked-attention mask transformer for universal image segmenta- tion,in:2022IEEE/CVFConferenceonComputerVisionandPattern Recognition (CVPR), pp

    Cheng, B., Misra, I., Schwing, A.G., Kirillov, A., Girdhar, R., 2022. Masked-attention mask transformer for universal image segmenta- tion,in:2022IEEE/CVFConferenceonComputerVisionandPattern Recognition (CVPR), pp. 1280–1289. doi:10.1109/CVPR52688.2022. 00135

    doi:10.1109/cvpr52688.2022 2022

  5. [5]

    Thecityscapesdataset forsemanticurbansceneunderstanding,in:2016IEEEConferenceon Computer Vision and Pattern Recognition (CVPR), pp

    Cordts,M.,Omran,M.,Ramos,S.,Rehfeld,T.,Enzweiler,M.,Benen- son,R.,Franke,U.,Roth,S.,Schiele,B.,2016. Thecityscapesdataset forsemanticurbansceneunderstanding,in:2016IEEEConferenceon Computer Vision and Pattern Recognition (CVPR), pp. 3213–3223. doi:10.1109/CVPR.2016.350

    doi:10.1109/cvpr.2016.350 2016

  6. [6]

    Carla: An open urban driving simulator.arXiv:1711.03938

    Dosovitskiy, A., Ros, G., Codevilla, F., Lopez, A., Koltun, V., 2017. Carla: An open urban driving simulator.arXiv:1711.03938

    Pith/arXiv arXiv 2017

  7. [7]

    The faiss li- brary.IEEETransactionsonBigData,1–17doi:10.1109/TBDATA.2025

    Douze, M., Guzhva, A., Deng, C., Johnson, J., Szilvasy, G., Mazaré, P.E., Lomeli, M., Hosseini, L., Jégou, H., 2025. The faiss li- brary.IEEETransactionsonBigData,1–17doi:10.1109/TBDATA.2025. 3618474

    work page doi:10.1109/tbdata.2025 2025

  8. [8]

    Towards pragmatic semantic image synthesis for urban scenes, in: 2023 IEEE Intelligent Vehicles Symposium (IV), pp

    Eskandar, G., Guo, D., Guirguis, K., Yang, B., 2023. Towards pragmatic semantic image synthesis for urban scenes, in: 2023 IEEE Intelligent Vehicles Symposium (IV), pp. 1–8. doi:10.1109/IV55152. 2023.10186767

    doi:10.1109/iv55152 2023

  9. [9]

    Dual contrastive learning for unsupervised image-to-image translation, in: 2021IEEE/CVFConferenceonComputerVisionandPatternRecog- nition Workshops (CVPRW), pp

    Han, J., Shoeiby, M., Petersson, L., Armin, M.A., 2021. Dual contrastive learning for unsupervised image-to-image translation, in: 2021IEEE/CVFConferenceonComputerVisionandPatternRecog- nition Workshops (CVPRW), pp. 746–755. doi:10.1109/CVPRW53098. 2021.00084

    doi:10.1109/cvprw53098 2021

  10. [10]

    Hoyez, H., Schockaert, C., Rambach, J., Mirbach, B., Stricker, D.,

  11. [11]

    Unsupervised image-to-image translation: A review. Sensors

  12. [12]

    doi:10.3390/s22218540

    work page doi:10.3390/s22218540

  13. [13]

    Toward real-time g-buffer-guided style transfer in computer games

    Ioannou, E., Maddock, S., 2025. Toward real-time g-buffer-guided style transfer in computer games. IEEE Transactions on Games 17, 613–621. doi:10.1109/TG.2024.3372829

    work page doi:10.1109/tg.2024.3372829 2025

  14. [14]

    Very deep convolutional neural network based image classification using small training sample size, in: 2015 3rd IAPR Asian Conference on Pattern Recognition (ACPR), pp

    Liu, S., Deng, W., 2015. Very deep convolutional neural network based image classification using small training sample size, in: 2015 3rd IAPR Asian Conference on Pattern Recognition (ACPR), pp. 730–734. doi:10.1109/ACPR.2015.7486599

    doi:10.1109/acpr.2015.7486599 2015

  15. [15]

    Least squares generative adversarial networks, in: 2017 IEEE Inter- national Conference on Computer Vision (ICCV), pp

    Mao, X., Li, Q., Xie, H., Lau, R.Y., Wang, Z., Smolley, S.P., 2017. Least squares generative adversarial networks, in: 2017 IEEE Inter- national Conference on Computer Vision (ICCV), pp. 2813–2821. doi:10.1109/ICCV.2017.304

    doi:10.1109/iccv.2017.304 2017

  16. [16]

    The mapillary vistas dataset for semantic understanding of street scenes, in:2017IEEEInternationalConferenceonComputerVision(ICCV), pp

    Neuhold, G., Ollmann, T., Bulò, S.R., Kontschieder, P., 2017. The mapillary vistas dataset for semantic understanding of street scenes, in:2017IEEEInternationalConferenceonComputerVision(ICCV), pp. 5000–5009. doi:10.1109/ICCV.2017.534

    doi:10.1109/iccv.2017.534 2017

  17. [17]

    Effects of sim2realimagetranslationviadclganonlanekeepingassistsystemin carla simulator

    Pahk, J., Shim, J., Baek, M., Lim, Y., Choi, G., 2023. Effects of sim2realimagetranslationviadclganonlanekeepingassistsystemin carla simulator. IEEE Access 11, 33915–33927. doi:10.1109/ACCESS. 2023.3262991

    doi:10.1109/access 2023

  18. [18]

    Contrastivelearning for unpaired image-to-image translation.arXiv:2007.15651

    Park,T.,Efros,A.A.,Zhang,R.,Zhu,J.Y.,2020. Contrastivelearning for unpaired image-to-image translation.arXiv:2007.15651

    Pith/arXiv arXiv 2020

  19. [19]

    Photorealistic synthetic crowds simulation in indoor environments (pscs-i): a novel synthetic dataset for realistic simulation of crowd panic and violence behaviors

    Pasios, S., Gkountakos, K., Ioannidis, K., Tsikrika, T., Vrochidis, S., Kompatsiaris, I., 2025. Photorealistic synthetic crowds simulation in indoor environments (pscs-i): a novel synthetic dataset for realistic simulation of crowd panic and violence behaviors. IEEE Access , 1–1doi:10.1109/ACCESS.2025.3636727

    work page doi:10.1109/access.2025.3636727 2025

  20. [20]

    Carla2real: A tool for reducing the sim2real appearance gap in carla simulator

    Pasios, S., Nikolaidis, N., 2025a. Carla2real: A tool for reducing the sim2real appearance gap in carla simulator. IEEE Transactions on Intelligent Transportation Systems 26, 18747–18761. doi:10.1109/ TITS.2025.3597010

    arXiv 2025

  21. [21]

    Enhancing photorealism in carla autonomous driving simulator, in: 2025 33rd European Signal Processing Conference (EUSIPCO), pp

    Pasios, S., Nikolaidis, N., 2025b. Enhancing photorealism in carla autonomous driving simulator, in: 2025 33rd European Signal Processing Conference (EUSIPCO), pp. 671–675. doi:10.23919/ EUSIPCO63237.2025.11226280

    arXiv 2025

  22. [22]

    Regen: Real-time photorealism en- hancement in games via a dual-stage generative network framework

    Pasios, S., Nikolaidis, N., 2026. Regen: Real-time photorealism en- hancement in games via a dual-stage generative network framework. IEEE Transactions on Games , 1–8doi:10.1109/TG.2026.3661622

    work page doi:10.1109/tg.2026.3661622 2026

  23. [23]

    Enhancing photoreal- ism enhancement

    Richter, S.R., Alhaija, H.A., Koltun, V., 2023. Enhancing photoreal- ism enhancement. IEEE Transactions on Pattern Analysis and Ma- chine Intelligence 45, 1700–1715. doi:10.1109/TPAMI.2022.3166687

    doi:10.1109/tpami.2022.3166687 2023

  24. [24]

    Playing for benchmarks, in: IEEE International Conference on Computer Vision, ICCV 2017, Venice,Italy,October22-29,2017,pp.2232–2241.doi:10.1109/ICCV

    Richter, S.R., Hayder, Z., Koltun, V., 2017. Playing for benchmarks, in: IEEE International Conference on Computer Vision, ICCV 2017, Venice,Italy,October22-29,2017,pp.2232–2241.doi:10.1109/ICCV. 2017.243

    doi:10.1109/iccv 2017

  25. [25]

    Playingfordata: Ground truth from computer games.arXiv:1608.02192

    Richter,S.R.,Vineet,V.,Roth,S.,Koltun,V.,2016. Playingfordata: Ground truth from computer games.arXiv:1608.02192

    Pith/arXiv arXiv 2016

  26. [26]

    U-net: Convolutional networksforbiomedicalimagesegmentation,in:Navab,N.,Horneg- ger, J., Wells, W.M., Frangi, A.F

    Ronneberger, O., Fischer, P., Brox, T., 2015. U-net: Convolutional networksforbiomedicalimagesegmentation,in:Navab,N.,Horneg- ger, J., Wells, W.M., Frangi, A.F. (Eds.), Medical Image Computing andComputer-AssistedIntervention–MICCAI2015,SpringerInter- national Publishing, Cham. pp. 234–241

    2015

  27. [27]

    Shaham, T.R., Gharbi, M., Zhang, R., Shechtman, E., Michaeli, T.,

  28. [28]

    Spatially-adaptive pixelwise networks for fast image transla- tion,in:2021IEEE/CVFConferenceonComputerVisionandPattern Recognition(CVPR),pp.14877–14886.doi:10.1109/CVPR46437.2021. 01464

    doi:10.1109/cvpr46437.2021 2021

  29. [29]

    Unpaired image translation via vector symbolic architectures, in: European Conference on Computer Vision

    Theiss, J.D., Leverett, J., Kim, D., Prakash, A., 2022. Unpaired image translation via vector symbolic architectures, in: European Conference on Computer Vision

    2022

  30. [30]

    Yolov12: Attention-centric real-time object detectors

    Tian, Y., Ye, Q., Doermann, D., 2025. Yolov12: Attention-centric real-time object detectors. arXiv preprint arXiv:2502.12524

    Pith/arXiv arXiv 2025

  31. [31]

    Wang, T.C., Liu, M.Y., Zhu, J.Y., Tao, A., Kautz, J., Catanzaro, B.,

  32. [32]

    8798–8807

    High-resolution image synthesis and semantic manipulation with conditional gans, in: 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 8798–8807. doi:10.1109/CVPR. 2018.00917

    doi:10.1109/cvpr 2018

  33. [33]

    Adding conditional control to text-to-image diffusion models, in: 2023 IEEE/CVF International Conference on Computer Vision (ICCV), pp

    Zhang, L., Rao, A., Agrawala, M., 2023. Adding conditional control to text-to-image diffusion models, in: 2023 IEEE/CVF International Conference on Computer Vision (ICCV), pp. 3813–3824. doi:10. 1109/ICCV51070.2023.00355. S. Pasios et al.:Preprint submitted to ElsevierPage 7 of 7

    arXiv 2023