Hu, Teng, Yi, Ran, Huang, Hongrui, Ma, Lizhuang, Rosin, Paul L.  ORCID: https://orcid.org/0000-0002-4965-3884 and Lai, Yukun ORCID: https://orcid.org/0000-0002-2094-5680
2026.
SuperSVG++: Semantic-driven layer-wise scalable vector graphics synthesis with semantic layer decomposition.
International Journal of Computer Vision
134
(11)
, 497.
10.1007/s11263-026-03099-1
|
Preview |
PDF
- Accepted Post-Print Version
Available under License Creative Commons Attribution. Download (16MB) | Preview |
Abstract
Scalable Vector Graphics (SVG) is a widely used format due to its excellent scalability and editability. However, existing image vectorization methods either suffer from poor reconstruction quality for complex images or require excessive computation time, limiting their practical application. To address these challenges, we propose SuperSVG++, a semantic-driven, layer-wise vectorization model that converts raster images into SVG with high accuracy and efficiency. Firstly, SuperSVG++ proposes a semantic-aware hierarchical vectorization framework, which extracts and reorders semantic layers in a depth-aware manner, ensuring structured and semantic-aware vectorization. Furthermore, we propose a soft-hard boundary constraint in the semantic-aware coarse structure vectorization stage, which improves both the compactness and accuracy of vector paths. To refine details, we propose a difference-conditioned refinement model, which adaptively predicts refining paths based on a difference map computed between the rendered and target image. Additionally, we propose a layer-wise refinement strategy, utilizing a distance-aware priority queue to dynamically decompose and refine semantic regions in a coarse-to-fine and layer-wise manner. Extensive experiments demonstrate that SuperSVG++ outperforms state-of-the-art methods in both reconstruction accuracy and semantic consistency, achieving superior layer-wise vectorization performance.
| Item Type: | Article |
|---|---|
| Date Type: | Publication |
| Status: | Published |
| Schools: | Schools > Computational & Mathematical Sciences Schools > Computer Science & Informatics |
| Additional Information: | RRS policy appplied |
| Publisher: | Springer |
| ISSN: | 0920-5691 |
| Date of First Compliant Deposit: | 4 November 2026 |
| Date of Acceptance: | 5 October 2026 |
| Last Modified: | 04 Nov 2026 10:15 |
| URI: | https://orca.cardiff.ac.uk/id/eprint/190597 |
Actions (repository staff only)
 |
Edit Item |
Dimensions
Dimensions