Research map / Image generation / Image restoration
Editable Image Geometric Abstraction via Neural Primitive Assembly
Ye Chen, Bingbing Ni, Xuanhong Chen, Zhangli Hu
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 neural soft assignment (opacity-weighted primitive selection) instead of hard argmax assignment (ablation) | primitive assignment · Emoji | training MSE | Improved |
| use neural soft assignment instead of HardAssign+Gumbel Softmax (ablation) | primitive assignment · Emoji | training MSE | Improved |
| add centroid filtering with Gaussian kernel (ablation) | full method · Emoji | image loss / fit | Improved |
| add annealing of the Gaussian sigma (ablation) | centroid filtering · Emoji | image loss / blur | Improved |
| add point-wise offsets to affine transformation (ablation) | full method · Emoji | MSE | Improved |
| use too large initial sigma or too small annealing rate (ablation) | centroid filtering · Emoji | MSE | Hurt |
| vary Gaussian kernel size (ablation) | centroid filtering · Emoji | MSE: all MSEs below 0.003 | No clear effect |
| reduce number of shapes to 5 (ablation) | full method · Emoji | MSE | No clear effect |
| use geometric primitives (4 types) instead of sketch strokes or triangles only | image abstraction · Bauhaus/Icon/Emoji/Emoji-Aug | geometric distance, retrieval accuracy: Icon retrieval 51.85 -> 77.78 vs CLIPasso | Improved |
| primitive-based method vs LIVE Bezier vectorization | image vectorization · Emoji / Emoji-Aug | pixel MSE: Emoji 0.0016 (LIVE) vs 0.0020 (ours) | Mixed |
Closest papers by meaning
- Learning Shape Abstractions by Assembling Volumetric Primitives (CVPR 2017 · similarity 0.85)
- Towards High-fidelity Artistic Image Vectorization via Texture-Encapsulated Shape Parameterization (CVPR 2024 · similarity 0.84)
- Less is More: Efficient Image Vectorization with Adaptive Parameterization (CVPR 2025 · similarity 0.84)
- Optimize & Reduce: A Top-Down Approach for Image Vectorization (AAAI 2024 · similarity 0.83)
- Image Sculpting: Precise Object Editing with 3D Geometry Control (CVPR 2024 · similarity 0.83)
- Training-free Spatially Grounded Geometric Shape Encoding (NeurIPS 2026 · similarity 0.83)
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.
