Pith. sign in

REVIEW 5 major objections 5 minor 39 references

Vectorized Region Based Brush Strokes for Artistic Rendering

T0 review · 5 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Pith's one-line read Automatic segmentation plus SVG vectorization turns any image into a region-guided brush stroke sequence.

desk verdict A plausible, training-free region-guided stroke pipeline that deserves peer review but must not be accepted with its current qualitative-only evidence and a concrete sort-order specification bug. read the letter →

arxiv 2506.09969 v1 pith:OXHWNIPO submitted 2025-06-11 cs.CV

classification cs.CV
keywords stroke-basedrenderingsemanticsegmentationSVGvectorizationbrushstrokesynthesispaintingevolutionregion-drivenparametricstrokesimage-to-painting
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 proposes a stroke-based rendering method that decides both where and what to paint by combining automatic image segmentation with SVG vectorization. The authors argue that decomposing an image into semantic segments, then further into vectorized bezier patches, builds a region hierarchy that lets brush strokes follow objects and parts in a natural order. Stroke parameters—position, size, angle, and color—are derived from minimum rotated rectangles fitted to each polygonized vector region, and a proximity-based sequencing step orders the regions and strokes. On face, painting, photographic, and abstract inputs, the method claims to render a faithful painting stroke by stroke, region by region, and segment by segment, aligning with a region-based painting strategy.

What carries the argument

The central mechanism is the hierarchical region pipeline: SAM automatic mask generation with IoU-based filtering produces non-overlapping semantic segments; SVG vectorization converts each segment into bezier patches (lines, quadratic/cubic bezier curves, circular and elliptical arcs); each patch is polygonized and, for large patches, grid-decomposed; and the minimum rotated rectangle of each sub-polygon supplies the stroke parameter tuple $(x, y, w, h, \theta, r, g, b)$. A proximity-based sequence generator, using hierarchical clustering and the traveling salesman problem, orders the vectorized regions, and $\alpha$ blending composites each stroke onto the evolving canvas.

What would settle it

Run SAM's automatic mask generator with the paper's IoU filtering on an image of a single textured object and check whether the filtered segments are mutually disjoint and each aligns with one semantic part; any overlapping or fragmented segment that splits a coherent object breaks the claimed region-driven semantic order of strokes.

Watch

Extended reading notes

Core claim

The central claim is that a scene-to-segment-to-SVG-region-to-stroke hierarchy, built from SAM automatic segmentation and SVG vectorization, produces brush stroke sequences that follow semantic regions while preserving image fidelity. Unlike grid-based or pixel-loss-driven painters, the method treats each segment as a painting region, subdivides it into vectorized bezier patches, approximates each patch by a minimum rotated rectangle to obtain stroke geometry, and orders strokes by perceptual proximity. The paper reports that this reproduces the way an artist paints—background first, then larger regions, then details—while supporting arbitrary resolutions and aspect ratios. The result is a stroke-by-stroke painting evolution that is both high fidelity and interpretable.

Load-bearing premise

The method assumes that SAM's automatic masks, after IoU filtering, are non-overlapping and correspond to the semantic regions an artist would choose to paint separately.

Editorial extensions

If this is right

  • The method yields a stroke-by-stroke evolution video whose order is dictated by semantic regions, which can support art education and virtual exhibitions.
  • Because it avoids pixel-loss optimization and grid-based region prediction, the method handles arbitrary image resolutions and aspect ratios without blurring.
  • Each stroke is tied to a specific semantic region, so editing or retouching a stroke corresponds to editing a localized region of the image.
  • The same pipeline applies to photographs, face images, paintings, style-transferred images, and abstract art with no retraining.

Reading between the lines

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

  • If the semantic-region assumption holds, the generated stroke order could be compared against human artists' painting order (e.g., from recorded painting videos) to give a quantitative test of region-based painting strategies.
  • Replacing SAM with another segmentation model would reveal how much of the stroke-ordering benefit comes from the segmenter rather than from the vectorization and sequencing machinery.
  • The grid decomposition parameter offers a direct user control for stroke density, suggesting a way to trade fidelity against painting time without changing the region hierarchy.
  • The method's quality is bounded by SVG vectorization fidelity; increasing the number of control points per patch would likely sharpen strokes at the cost of more strokes per region.
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

5 major / 5 minor

Summary. The paper proposes a stroke-based rendering (SBR) pipeline that combines SAM segmentation, SVG vectorization, polygon approximation, and alpha blending. The method generates a stroke-by-stroke painting sequence following a scene-to-segment-to-SVG-region-to-stroke hierarchy, with the goal of aligning with region-based artistic strategies. Experiments are reported on WikiArt, METFace, FFHQ, Im2Oil, Delaunay, and style-transferred images, with qualitative comparisons against SNP, CNP, sketch & paint, and SBPP. The evaluation is entirely qualitative, with no quantitative metrics or user studies.

Significance. Strengths: the pipeline is modular and uses publicly available components (SAM, Shapely), and the provided videos offer a concrete demonstration of the intended stroke-by-stroke behavior. If supported, the hierarchical organization addresses a recognized limitation in SBR. However, the central claims of 'high fidelity and superior stroke quality' are empirical and comparative, and the paper currently provides no measurable evidence for them. The contribution is primarily an engineered combination of existing tools rather than a new theoretical result; with a stronger evaluation this could be acceptable for an applied venue.

major comments (5)
  1. [Abstract; Section 3.3] The abstract claims 'high fidelity and superior stroke quality', which is an empirical comparative claim, but Section 3.3 explicitly states that the evaluation is qualitative. No quantitative metrics (e.g., PSNR, LPIPS, FID, CLIP score), no stroke-count or runtime comparisons, and no perceptual study are reported. As written, the headline claim is not testable. Please add quantitative comparisons against the cited baselines at matched stroke budgets, or a user study, or temper the claim to the qualitative evidence actually provided.
  2. [Section 2.1, Eq. (1)] The mask-refinement step is under-specified and internally inconsistent: the text says masks are sorted by area 'low to high' for IoU filtering, while Eq. (1) says 'sorting based on area (high to low)'. Since the output of IoU-based non-overlapping filtering depends on processing order, the segment set is not uniquely reproducible from the paper. Please specify the exact algorithm, including the IoU threshold, tie-breaking, and whether the sort order is ascending or descending, and release code or pseudo-code for this step.
  3. [Section 2.2] The hyperparameters pgrid, pgroup, and the area threshold δ are introduced without a concrete rule or sensitivity analysis. The stroke count, geometry, and final appearance all depend on these values, and the text only says they are 'configured based on the size of the polygon'. This makes the method difficult to reproduce and obscures which components drive the qualitative results. Provide explicit formulas or a sensitivity study (e.g., final metrics or stroke count as a function of pgrid and pgroup).
  4. [Section 2.2 vs. Section 3.2] The stroke geometry is described inconsistently: Section 2.2 says 'minimum rotated rectangular polygon', while Section 3.2 says 'Shapely minimum-rotated triangle'. These yield different stroke shapes and affect the parameter estimates for position and angle. Please correct the inconsistency and specify the exact Shapely function and its parameters.
  5. [Section 2.2, Eq. (4)] Equation (4) does not appear to be standard alpha compositing. With A_overlay = 1, the formula gives C_result = C_base, so the overlay (stroke) color never replaces the base; with intermediate alpha, the base is darkened by multiplication with the overlay color. If the intended operation is source-over compositing, the formula should be C_result = C_overlay * A_overlay + C_base * (1 - A_overlay). Please clarify the intended compositing operator and justify why the stated formula produces faithful rendering.
minor comments (5)
  1. [Section 3.2] Please describe how the SAM parameters ('points per side' from 2 to 8, IoU and stability thresholds between 0.6 and 0.8) are selected for each input image, and whether this selection affects the semantic validity of the resulting segments.
  2. [Figure 4] The qualitative comparison would be more informative with zoomed crops at matched resolution and with a table of stroke counts, resolutions, and runtimes for each method and image.
  3. [References [21], [25], [26]] The sequencing method 'Seq Gen' is attributed to the authors' own prior work [21]; please make the description in Eq. (3) self-contained or point to the specific subsection of [21] that defines the algorithm. For the YouTube playlists, add access dates and note that they are supplementary materials.
  4. [Global] There are minor language issues, e.g., 'We detail out the data collection' in Section 3.1 and inconsistent spacing in 'B ´ezier'. A careful proofreading pass is recommended.
  5. [Conclusion] The paper would benefit from an explicit limitations paragraph, e.g., regarding the dependence on SAM's mask quality, the lack of quantitative validation, and the computational cost of vectorization and polygon processing.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; the paper is an engineering pipeline whose stroke parameters are extracted from the input by construction, and its headline fidelity claim is qualitatively asserted but not derived from a fitted or self-cited reduction.

full rationale

The derivation chain is transparent: SAM segmentation (Eq. 1), SVG vectorization (Eq. 2), sequencing (Eq. 3), and alpha-blending rendering (Eq. 4). Stroke positions and colors are computed deterministically from the input's vectorized regions, so high reconstruction similarity is unsurprising by design, but this is not a circular prediction: the paper does not fit a parameter to a subset and then claim to predict that subset. The only self-citation, Seq Gen [21], supplies an ordering component and is not used to justify the headline claim of region-based stroke quality; that claim is supported only by qualitative figures. Section 3.3 explicitly states: "Since there is no formal quantitative measure to evaluate the stroke-by-stroke painting process..., we primarily assess the effectiveness of our approach qualitatively," which is an evidentiary gap, not a circular reduction. The internal inconsistency in Section 2.1 (sorting masks low-to-high versus high-to-low) makes the preprocessing under-specified but does not make the derivation circular. No central claim reduces by construction to its own input or to a self-citation chain.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The central claim relies on semantic segmentation quality and vectorization fidelity, while the method's internal hyperparameters are manually set rather than fitted or learned. No new entities are introduced.

free parameters (4)
  • pgrid = not stated
    Grid decomposition parameter for large polygons, configured based on polygon size; controls stroke subdivision and is chosen by hand (Section 2.2).
  • pgroup = not stated
    Number of strokes per region; for large polygons it subdivides, for small polygons it equals the polygon itself; set manually and directly affects stroke density (Section 2.2).
  • segment area threshold delta = not stated
    Threshold deciding when grid decomposition is applied; not quantified in the paper (Section 2.2).
  • SAM hyperparameters = points per side 2-8, IoU threshold 0.6-0.8, stability score 0.6-0.8
    Chosen by hand in Section 3.2; these settings affect the number and quality of segments but are not justified per image.
assumptions (4)
  • domain assumption SAM's automatic mask generator, after IoU filtering, yields non-overlapping, semantically meaningful segments for arbitrary images
    Invoked in Section 2.1; the whole region-driven rendering depends on this, and no quantitative validation of segment quality is provided.
  • domain assumption Proximity-based hierarchical clustering plus TSP recovers a human-like painting order
    Section 2.2 cites Gestalt theory and prior work [21], but no user study or comparison against real artist painting sequences is performed.
  • domain assumption Minimum rotated rectangle approximation of polygon patches preserves the essential geometry for brush strokes
    Section 2.2 and 3.2 use Shapely minimum rotated triangles/rectangles; the approximation discards shape detail, and the paper does not quantify the resulting fidelity loss.
  • domain assumption The SVG vectorization tool accurately captures the structure of each segment
    Section 2.1 relies on a 'publicly available vectorization tool' without naming it or analyzing its errors; stroke quality depends on this representation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Vectorized Region Based Brush Strokes for Artistic Rendering." pith.science (2026). https://pith.science/paper/OXHWNIPO

@misc{pith2026250609969,
  author       = {Pith},
  title        = {Pith review of: Vectorized Region Based Brush Strokes for Artistic Rendering},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OXHWNIPO}},
  note         = {Machine review of arXiv:2506.09969}
}
read the original abstract

Creating a stroke-by-stroke evolution process of a visual artwork tries to bridge the emotional and educational gap between the finished static artwork and its creation process. Recent stroke-based painting systems focus on capturing stroke details by predicting and iteratively refining stroke parameters to maximize the similarity between the input image and the rendered output. However, these methods often struggle to produce stroke compositions that align with artistic principles and intent. To address this, we explore an image-to-painting method that (i) facilitates semantic guidance for brush strokes in targeted regions, (ii) computes the brush stroke parameters, and (iii) establishes a sequence among segments and strokes to sequentially render the final painting. Experimental results on various input image types, such as face images, paintings, and photographic images, show that our method aligns with a region-based painting strategy while rendering a painting with high fidelity and superior stroke quality.

Figures

Figures reproduced from arXiv: 2506.09969 by the authors.

Figure 1
Figure 1. Our painting process involves (i) dividing the input image into a set of pragmatic regions using segmentation and vectorization, (ii) [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Stroke-by-stroke painting evolution on WikiArt samples. [Zoom in to observe stroke details closely] [The complete painting [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Illustration of Stroke-by-stroke painting evolution on a) METFace [ [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Qualitative comparison of our proposed method with a) Compositional neural painter (CNP) [ [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

39 extracted references · 33 canonical work pages

  1. [1]

    End- toend object detection with transformers

    Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, and Sergey Zagoruyko. End- toend object detection with transformers. in eccv. Springer, 1(2):4, 2020. 1

  2. [2]

    The genesis of errors in drawing

    Rebecca Chamberlain and Johan Wagemans. The genesis of errors in drawing. Neuroscience & Biobehavioral Reviews, 65:195–207, 2016. 1

  3. [3]

    Wetbrush: Gpu-based 3d painting simulation at the bristle level

    Zhili Chen, Byungmoon Kim, Daichi Ito, and Huamin Wang. Wetbrush: Gpu-based 3d painting simulation at the bristle level. ACM Transactions on Graphics (TOG) , 34(6):1–11,

  4. [4]

    Look little, look often: The influence of gaze frequency on drawing accuracy

    Dale J Cohen. Look little, look often: The influence of gaze frequency on drawing accuracy. Perception & psy- chophysics, 67:997–1009, 2005. 1

  5. [5]

    Segmentation-based parametric painting

    Manuel Ladron de Guevara, Matt Fisher, and Aaron Hertz- mann. Segmentation-based parametric painting. In 2024 IEEE International Conference on Multimedia and Expo Workshops (ICMEW), pages 1–6. IEEE, 2024. 1, 4, 7

  6. [6]

    Algo- rithms for the decomposition of a polygon into convex poly- gons

    Jos ´e Fern ´andez, L ´azaro C ´anovas, and B Pelegrın. Algo- rithms for the decomposition of a polygon into convex poly- gons. European Journal of Operational Research , 121(2): 330–342, 2000. 3

  7. [7]

    Style transfer - replicate, 2025

    Fofr. Style transfer - replicate, 2025. 7

  8. [8]

    Shapely: manipulation and analysis of geometric objects

    Sean Gillies and contributors. Shapely: manipulation and analysis of geometric objects. GitHub repository, 2007– present. 3

Show all 39 references
  1. [9]

    Visual and motor processing in visual artists: Implications for cognitive and neural mechanisms

    Kuba Glazek. Visual and motor processing in visual artists: Implications for cognitive and neural mechanisms. Psychol- ogy of Aesthetics, Creativity, and the Arts , 6(2):155, 2012. 1

  2. [10]

    De- launay: a dataset of abstract art for psychophysical and ma- chine learning research

    Camille Gontier, Jakob Jordan, and Mihai A Petrovici. De- launay: a dataset of abstract art for psychophysical and ma- chine learning research. arXiv preprint arXiv:2201.12123 ,

  3. [11]

    Stroke- based neural painting and stylization with dynamically pre- dicted painting region

    Teng Hu, Ran Yi, Haokun Zhu, Liang Liu, Jinlong Peng, Yabiao Wang, Chengjie Wang, and Lizhuang Ma. Stroke- based neural painting and stylization with dynamically pre- dicted painting region. In Proceedings of the 31st ACM International Conference on Multimedia, pages 7470–7480,

  4. [12]

    Towards artist-like painting agents with multi-granularity semantic alignment

    Zhangli Hu, Ye Chen, Zhongyin Zhao, Jinfan Liu, Bilian Ke, and Bingbing Ni. Towards artist-like painting agents with multi-granularity semantic alignment. In Proceedings of the 32nd ACM International Conference on Multimedia , pages 10191–10199, 2024. 1, 3

  5. [13]

    Learning to paint with model-based deep reinforcement learning

    Zhewei Huang, Wen Heng, and Shuchang Zhou. Learning to paint with model-based deep reinforcement learning. In Proceedings of the IEEE/CVF international conference on computer vision, pages 8709–8718, 2019. 1

  6. [14]

    A style-based generator architecture for genera- tive adversarial networks

    Tero Karras. A style-based generator architecture for genera- tive adversarial networks. arXiv preprint arXiv:1812.04948,

  7. [15]

    Training generative adver- sarial networks with limited data

    Tero Karras, Miika Aittala, Janne Hellsten, Samuli Laine, Jaakko Lehtinen, and Timo Aila. Training generative adver- sarial networks with limited data. Advances in neural infor- mation processing systems, 33:12104–12114, 2020. 3, 7

  8. [16]

    Segment any- thing

    Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer White- head, Alexander C Berg, Wan-Yen Lo, et al. Segment any- thing. In Proceedings of the IEEE/CVF International Con- ference on Computer Vision, pages 4015–4026, 2023. 1, 2

  9. [17]

    A century of gestalt psychology in visual perception

    M Kubovy, SE Palmer, MA Peterson, M Singh, and R von der Heydt. A century of gestalt psychology in visual perception. Psychological Bulletin, 138:6, 2012. 2

  10. [18]

    Paint transformer: Feed forward neural painting with stroke prediction

    Songhua Liu, Tianwei Lin, Dongliang He, Fu Li, Ruifeng Deng, Xin Li, Errui Ding, and Hao Wang. Paint transformer: Feed forward neural painting with stroke prediction. In Pro- ceedings of the IEEE/CVF international conference on com- puter vision, pages 6598–6607, 2021. 1, 3

  11. [19]

    Stroke- based rendering: From heuristics to deep learning

    Florian Nolte, Andrew Melnik, and Helge Ritter. Stroke- based rendering: From heuristics to deep learning. arXiv preprint arXiv:2302.00595, 2022. 1

  12. [20]

    Drawing experts have better visual memory while drawing

    Florian Perdreau and Patrick Cavanagh. Drawing experts have better visual memory while drawing. Journal of Vision, 15(5):5–5, 2015. 1

  13. [21]

    Sketch & paint: Stroke-by-stroke evolution of visual artworks.arXiv preprint arXiv:2502.20119, 2025

    Jeripothula Prudviraj and Vikram Jamwal. Sketch & paint: Stroke-by-stroke evolution of visual artworks.arXiv preprint arXiv:2502.20119, 2025. 3, 4, 7

  14. [22]

    Sketching by perceptual grouping

    Yonggang Qi, Jun Guo, Yi Li, Honggang Zhang, Tao Xiang, and Yi-Zhe Song. Sketching by perceptual grouping. In2013 IEEE International Conference on Image Processing, pages 270–274. IEEE, 2013. 3

  15. [23]

    Learning a classification model for segmen- tation

    Ren and Malik. Learning a classification model for segmen- tation. In Proceedings ninth IEEE international conference on computer vision, pages 10–17. IEEE, 2003. 2

  16. [24]

    Large-scale classifica- tion of fine-art paintings: Learning the right metric on the right feature

    Babak Saleh and Ahmed Elgammal. Large-scale classifica- tion of fine-art paintings: Learning the right metric on the right feature. arXiv preprint arXiv:1505.00855, 2015. 3

  17. [25]

    Video of our vector painter pro- cess on datasets: Metface, ffhq, and im2oil

    Other Dataset Self. Video of our vector painter pro- cess on datasets: Metface, ffhq, and im2oil. https : //www.youtube.com/playlist?list=PLDykw6_ Rz8y9zog3wPrqZiri2bt4WCtUB, 2025. 7

  18. [26]

    Video of our vector painter process on wikiart dataset

    Wikiart Self. Video of our vector painter process on wikiart dataset. https://www.youtube.com/playlist? list=PLDykw6_Rz8y85MwZ8HcyU91OIkjepw62- ,

  19. [27]

    Intelli-paint: Towards developing more human- intelligible painting agents

    Jaskirat Singh, Cameron Smith, Jose Echevarria, and Liang Zheng. Intelli-paint: Towards developing more human- intelligible painting agents. In European Conference on Computer Vision, pages 685–701. Springer, 2022. 1

  20. [28]

    Proces- spainter: Learn painting process from sequence data

    Yiren Song, Shijie Huang, Chen Yao, Xiaojun Ye, Hai Ci, Jiaming Liu, Yuxuan Zhang, and Mike Zheng Shou. Proces- spainter: Learn painting process from sequence data. arXiv preprint arXiv:2406.06062, 2024. 1

  21. [29]

    Free hand-drawn sketch segmentation

    Zhenbang Sun, Changhu Wang, Liqing Zhang, and Lei Zhang. Free hand-drawn sketch segmentation. In Computer Vision–ECCV 2012: 12th European Conference on Com- puter Vision, Florence, Italy, October 7-13, 2012, Proceed- ings, Part I 12, pages 626–639. Springer, 2012. 3

  22. [30]

    Exploring the educational impact of diverse technologies in online virtual museums

    Stella Sylaiou, Katerina Mania, Ioannis Paliokas, Laia Pujol- Tost, Vassilis Killintzis, and Fotis Liarokapis. Exploring the educational impact of diverse technologies in online virtual museums. International Journal of Arts and Technology, 10 (1):58–84, 2017. 1

  23. [31]

    Segmentation and accuracy in copying and drawing: Experts and beginners

    John Tchalenko. Segmentation and accuracy in copying and drawing: Experts and beginners. Vision research, 49(8):791– 800, 2009. 1

  24. [32]

    Im2oil: stroke-based oil painting rendering with linearly controllable fineness via adaptive sampling

    Zhengyan Tong, Xiaohang Wang, Shengchao Yuan, Xuan- hong Chen, Junjie Wang, and Xiangzhong Fang. Im2oil: stroke-based oil painting rendering with linearly controllable fineness via adaptive sampling. In Proceedings of the 30th ACM International Conference on Multimedia, pages ...

  25. [33]

    A century of gestalt psychology in visual perception: I

    Johan Wagemans, James H Elder, Michael Kubovy, Stephen E Palmer, Mary A Peterson, Manish Singh, and R¨udiger V on der Heydt. A century of gestalt psychology in visual perception: I. perceptual grouping and figure–ground organization. Psychological bulletin, 138(6):1172, 2012. 2

  26. [34]

    Unsupervised im- age segmentation using convolutional autoencoder with total variation regularization as preprocessing

    Chunlai Wang, Bin Yang, and Yiwen Liao. Unsupervised im- age segmentation using convolutional autoencoder with total variation regularization as preprocessing. In 2017 IEEE In- ternational Conference on Acoustics, Speech and Signal Pro- cessing (ICASSP), pages 1877–1881. IEEE, 2017. 2

  27. [35]

    Poly- gon area decomposition using a compactness metric

    Mariusz Wzorek, Cyrille Berger, and Patrick Doherty. Poly- gon area decomposition using a compactness metric. arXiv preprint arXiv:2110.04043, 2021. 3

  28. [36]

    Depth anything: Unleashing the power of large-scale unlabeled data

    Lihe Yang, Bingyi Kang, Zilong Huang, Xiaogang Xu, Jiashi Feng, and Hengshuang Zhao. Depth anything: Unleashing the power of large-scale unlabeled data. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 10371–10381, 2024. 1

  29. [37]

    Pastiche master: Exemplar-based high-resolution por- trait style transfer

    Shuai Yang, Liming Jiang, Ziwei Liu, and Chen Change Loy. Pastiche master: Exemplar-based high-resolution por- trait style transfer. In CVPR, 2022. 7

  30. [38]

    From image parsing to painterly rendering.ACM Trans

    Kun Zeng, Mingtian Zhao, Caiming Xiong, and Song Chun Zhu. From image parsing to painterly rendering.ACM Trans. Graph., 29(1):2–1, 2009. 1

  31. [39]

    Stylized neural painting

    Zhengxia Zou, Tianyang Shi, Shuang Qiu, Yi Yuan, and Zhenwei Shi. Stylized neural painting. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 15689–15698, 2021. 1, 3, 4, 7 A. Appendix Figure 3. Illustration of Stroke-by-stroke painting...

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.