Evaluation of Randomization through Style Transfer for Enhanced Domain Generalization

Dustin Eisenhardt; Timothy Schaumlöffel; Alperen Kantarci; Gemma Roig

cs.CVApr 7, 2026

Evaluation of Randomization through Style Transfer for Enhanced Domain Generalization

Authors: Dustin Eisenhardt, Timothy Schaumlöffel, Alperen Kantarci, Gemma Roig

Organizations: German Cancer Research Center (DKFZ), Heidelberg, Germany · German Cancer Consortium (DKTK), partner site Frankfurt/Mainz, a partnership between DKFZ and UCT Frankfurt-Marburg, Frankfurt am Main, Germany · Goethe University Frankfurt, Department of Computer Science, Frankfurt am Main, Germany · The Hessian Center for Artificial Intelligence (hessian.AI), Darmstadt, Germany

Abstract

Deep learning models for computer vision often suffer from poor generalization when deployed in real-world settings, especially when trained on synthetic data due to the well-known Sim2Real gap. Despite the growing popularity of style transfer as a data augmentation strategy for domain generalization, the literature contains unresolved contradictions regarding three key design axes: the diversity of the style pool, the role of texture complexity, and the choice of style source. We present a systematic empirical study that isolates and evaluates each of these factors for driving scene understanding, resolving inconsistencies in prior work. Our findings show that (i) expanding the style pool yields larger gains than repeated augmentation with few styles, (ii) texture complexity has no significant effect when the pool is sufficiently large, and (iii) diverse artistic styles outperform domain-aligned alternatives. Guided by these insights, we derive StyleMixDG (Style-Mixing for Domain Generalization), a lightweight, model-agnostic augmentation recipe that requires no architectural modifications or additional losses. Evaluated on the GTAV $\rightarrow$ {BDD100k, Cityscapes, Mapillary Vistas} benchmark, StyleMixDG demonstrates consistent improvements over strong baselines, confirming that the empirically identified design principles translate into practical gains. The code will be released on GitHub.

Figures & tables

Panel of Figure 1 — Figure 1: Overview of StyleMixDG. Each source image is stylized offline using N randomly selected styles, yielding an N-times augmented dataset. During training, images are sampled with 80% probability from the augmented set and 20% from the original, followed by photometric distortion.

Panel of Table 1 — Table 1: Examples of Style Transfer Augmentation (STA). This table shows examples of applying STA to synthetic images using style images of varying complexity. The first row depicts the style images taken from the Painter by Numbers [ 22 ] dataset.

Variant	ResNet50				ResNet101
	BDD	CS	MV	Avg	BDD	CS	MV	Avg
Source-only (Ours)	$17.20$	$23.51$	$33.62$	$24.78$	$24.83$	$28.70$	$37.92$	$30.48$
StyleMixDG (Ours)	$35.61$	$40.75$	$42.45$	$39.54$	$36.34$	$42.74$	$45.15$	$41.41$
GTR [ 20 ]	$33.75^{+}$	$37.53^{+}$	$34.52^{+}$	$35.16^{+}$	$39.60$	$43.70$	$39.10$	$40.80$
SAN-SAW [ 19 ]	$37.34$	$39.75$	$41.86$	$39.65$	$41.20^{*}$	$45.30^{*}$	$40.80^{*}$	$42.40^{*}$
WildNet [ 15 ]	$38.42^{+}$	$44.62^{+}$	$46.09^{+}$	$43.04^{+}$	$41.70^{*}$	$45.80^{*}$	$47.10^{*}$	$44.90^{*}$

Table 2: State-of-the-Art Comparison. Shows the mIoU achieved using different networks on the three benchmark datasets (BDD, CS, MV) and their average. Source-only refers to standard training on synthetic data without any domain generalization technique applied. Methods above the gray divider rely solely on augmentation without modifying the model architecture or training pipeline; methods below require substantial architectural or loss-level changes. Results are either taken from [ 19 ] as denoted by + , from [ 18 ] as denoted by → , and otherwise from the original paper.

Panel of Figure 2 — Figure 2: Qualitative Results. This figure shows qualitative results of applying the R101 Source-only and StyleMixDG variants to images from the target datasets.

Panel of Figure 3 — Figure 3: Results using more variation in style transfer. Source: training without style-transfer augmentation. STA is the baseline Style Transfer Augmentation. It is compared against Photo-metric Distortion (PMD), multiple style augmentations per image (MultAug) and sampling style images from a large pool of styles (Infty). Shows the mIoU achieved using different networks on the three benchmark datasets (BDD, CS, MV) and their average.

Variant	BDD	CS	MV	Avg
Infty	$28.99\pm 0.14$	$33.91\pm 0.24$	$34.45\pm 1.01$	$32.45\pm 0.28$
Low	$28.11\pm 0.70$	$34.35\pm 0.25$	$33.83\pm 0.28$	$32.10\pm 0.17$
Medium	$28.46\pm 0.97$	$33.25\pm 0.83$	$34.81\pm 1.02$	$32.17\pm 0.76$
High	$29.22\pm 1.24$	$33.58\pm 1.64$	$33.84\pm 0.89$	$32.21\pm 0.94$

Table 3: Results using different texture complexities. Compares sampling style images from low, medium and high-complexity styles with no complexity-based selection (Infty). The number of styles in the pool is identical. Shows the mIoU achieved using different networks on the three benchmark datasets (BDD, CS, MV) and their average.

Panel of Figure 4 — Figure 4: Results using different style sources. Style sources include the Painter-by-Numbers (Infty), GTAV (Intra) and benchmark datasets (BDD, CS, MV). Mixed denotes that all benchmark datasets are used to sample styles. Shows the mIoU achieved using different networks on the three benchmark datasets (BDD, CS, MV) and their average. The unsupervised domain adaptation settings are marked with an asterisk.

Infty	Mixing	MultAug	PMD	BDD	CS	MV	Average
✗	✗	✗	✗	$17.07\pm 0.60$	$20.43\pm 0.38$	$31.87\pm 0.31$	$23.12\pm 0.32$
✓	✗	✗	✗	$28.99\pm 0.14$	$33.91\pm 0.24$	$34.45\pm 1.01$	$32.45\pm 0.28$
✓	✗	✗	✓	$29.22\pm 0.86$	$34.87\pm 0.89$	$34.58\pm 0.45$	$32.89\pm 0.35$
✓	✓	✗	✗	$29.13\pm 0.84$	$33.57\pm 1.02$	$\mathbf{36.19\pm 0.66}$	$32.97\pm 0.47$
✓	✗	✓	✗	$29.09\pm 0.50$	$34.91\pm 0.13$	$34.77\pm 0.24$	$32.92\pm 0.20$
✓	✓	✓	✗	$\mathit{29.81\pm 0.12}$	$\mathit{36.03\pm 0.39}$	$35.66\pm 0.65$	$\mathit{33.83\pm 0.30}$

Table 4: Results using combinations of previous methods. Infty: increase the number of styles to sample from, Mixing: mix non-stylized and stylized images, MultAug: use multiple styles per image, PMD: apply photo-metric distortion. Shows the mIoU achieved using different networks on the three benchmark datasets (BDD, CS, MV) and their average. The best and second-best entries are highlighted in bold and italics respectively.

Variant	ResNet50 UperNet				DeIT-S16 UperNet
	BDD	CS	MV	Avg	BDD	CS	MV	Avg
Source	$30.54$	$28.79$	$35.04$	$31.46$	$33.74$	$33.04$	$38.63$	$35.14$
PMD	$34.56$	$39.83$	$39.30$	$37.89$	$32.87$	$\mathbf{38.69}$	$37.80$	$36.45$
StyleMixDG	$\mathbf{38.23}$	$\mathbf{41.07}$	$\mathbf{42.75}$	$\mathbf{40.68}$	$\mathbf{39.52}$	$38.19$	$\mathbf{42.29}$	$\mathbf{40.00}$

Table 5: Architecture Comparison. Source: non-stylized dataset, PMD: photo-metric distortion is applied, StyleMixDG: our method. Shows the mIoU achieved using different networks on the three benchmark datasets (BDD, CS, MV) and their average. The best entries are highlighted in bold .

Appendix figures & tables1 asset

Supplementary material from the paper’s appendix.

Appendix

GTAV [ 21 ]	BDD [ 27 ]	CS [ 7 ]	MV [ 17 ]
road	road	road

Appendix

Table 6: Label Correspondences. Label correspondences used in our experiments. The Label column additionally shows the color for each class used in subsequent visualization.

Domain Generalization

Domain generalization develops models that perform reliably in new environments without training on target-domain data.

See all 369 papers

Label-Decoupled Style Augmentation for Domain Generalization in Multi-Label Remote Sensing Scene Classification
Jul 14, 2026Alaa Almouradi, Erchan AptoulaRemote Sensing Image Understanding Multi-Label Classification
Faculty of Engineering and Natural Sciences, Sabancı University, 34956 Istanbul, Türkiye
Simple, Safe, and Overlooked: Reclaiming Sustainable Domain Generalization with Statistical Color Matching
Aug 19, 2026Sebastian Doerrich, Francesco Di Salvo, Shyam Nandan Rai +2Data Augmentation Domain Generalization
xAILab Bamberg, University of Bamberg, Bamberg, Germany
SHED: Style-Homogenized Embedding Alignment for Domain Generalization
May 16, 2026Kai Gan, Tong WeiVLM Robustness VLM Adaptation
School of Computer Science and Engineering, Southeast University, Nanjing 210096, China · Key Laboratory of Computer Network and Information Integration (Southeast University), Ministry of Education, China

Jul 14, 2026J/K move · Enter open · S save

Label-Decoupled Style Augmentation for Domain Generalization in Multi-Label Remote Sensing Scene Classification

Alaa Almouradi, Erchan Aptoula

Faculty of Engineering and Natural Sciences, Sabancı University, 34956 Istanbul, Türkiye

Remote Sensing Image Understanding Multi-Label Classification Data Augmentation Domain Generalization Remote Sensing Scene Classification

Open paper AskPDF arXiv

Multi-label classification assigns several co-occurring labels to each aerial scene, yet deployed models often encounter data distributions different from their training. Feature-statistics augmentation such as MixStyle, EFDMix, and correlated style uncertainty improves generalization at low cost but perturbs channel statistics globally, treating each image as a single style; one class can then contaminate the augmentation of another. Domain generalization is understudied for multi-label remote sensing; no prior method or multi-source benchmark targets it. A label-decoupled augmentation framework is therefore proposed, confining style perturbation to label-specific regions. Per-label attention, obtained from a learnable module or from gradient class-activation maps, yields per-label feature statistics; these statistics are mixed with cross-domain samples that share present labels, under independent per-label coefficients, and features are recomposed by attention-weighted normalization. Three operators combined with two attention sources produce six variants, evaluated on a leave-one-domain-out benchmark from multi-label UCM, AID, and DFC15 over six shared labels. Averaged over three splits and five seeds, the best variant attains 71.5% mean average precision, exceeding empirical risk minimization by 5.0 points and the strongest global-statistics baseline by 1.3 points, with the largest gain on the hardest transfer (up to 7.7 points). Ablations indicate that spatial attention and refreshed localization maps are most influential. The framework adds at most 0.35% parameters, leaves inference unchanged, and appears to offer a generic, inexpensive upgrade path for multi-label statistics-based domain generalization. Code is available upon acceptance at https://github.com/Alaa-Almouradi/Style-Augmentation-Upgrade.

Abstract

Domain Generalization

Explore similar work

Label-Decoupled Style Augmentation for Domain Generalization in Multi-Label Remote Sensing Scene Classification

Simple, Safe, and Overlooked: Reclaiming Sustainable Domain Generalization with Statistical Color Matching

SHED: Style-Homogenized Embedding Alignment for Domain Generalization