Research map / Detection and segmentation / Open-set and few-shot detection
SemLayer: Semantic-aware Generative Segmentation and Layer Construction for Abstract Icons
Haiyang Xu, Ronghuan Wu, Li-Yi Wei, Nanxuan Zhao, Chenxi Liu, Cuong Nguyen, Zhuowen Tu, Zhaowen Wang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use colorization-based generative segmentation instead of finetuned SAM2* | SemLayer segmentation · Icon test set (48 icons) | PQ: 59.4 -> 76.1 | Improved |
| use generative segmentation instead of finetuned SAM2* | SemLayer segmentation · Icon test set (48 icons) | mIoU: 79.3 -> 84.3 | Improved |
| use generative segmentation instead of off-the-shelf SAM2 | SemLayer segmentation · Icon test set (48 icons) | mIoU: 51.1 -> 84.3 | Improved |
| apply nearest-neighbor refinement of unlabeled pixels | SemLayer segmentation · Icon test set (48 icons) | PQ: 76.1 -> 78.3 | Improved |
| apply nearest-neighbor refinement | SAM2* · Icon test set (48 icons) | PQ: 59.4 -> 78.0 | Improved |
| apply nearest-neighbor refinement | gpt-image-1 · Icon test set (48 icons) | mIoU: 25.4 -> 57.2 | Improved |
| use better segmentation inputs to fixed completion module | SemLayer · Icon test set (48 icons) | Chamfer Distance: 49.1 -> 46.6 | Improved |
| finetune pix2gestalt completion on icon data instead of MP3D finetuned | SemLayer completion · Icon test set (48 icons) | mIoU: 75.3 -> 85.2 | Improved |
| use generative diffusion priors instead of classification/regression baseline | SemLayer · Icon test set (48 icons) | segmentation mIoU: 46.6 -> 84.3 | Improved |
| use generative diffusion priors instead of classification/regression baseline | SemLayer · Icon test set (48 icons) | completion mIoU: 63.9 -> 85.2 | Improved |
| use pixel-based pipeline instead of LLMs on raw SVG code | Gemini-3 and GPT-5 · Icon test set (48 icons) | validity | Hurt |
| use completion model instead of MP3D for speed | SemLayer completion · Icon test set (48 icons) | seconds per icon: 49.27 -> 16.83 | Improved |
Closest papers by meaning
- Decomposition of Graphic Design with Unified Multimodal Model (ICML 2025 · similarity 0.86)
- DrawingsDreamer: A Unified Multi-View Engineering Drawings Generation Model (NeurIPS 2026 · similarity 0.85)
- LayerCraft: Enhancing Text-to-Image Generation with CoT Reasoning and Layered Object Integration (NeurIPS 2025 · similarity 0.85)
- Vector Grimoire: Codebook-based Shape Generation under Raster Image Supervision (ICML 2025 · similarity 0.84)
- Chat2SVG: Vector Graphics Generation with Large Language Models and Image Diffusion Models (CVPR 2025 · similarity 0.84)
- VCG-Bench: Towards A Unified Visual-Centric Benchmark for Structured Generation and Editing (ICML 2026 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
