Research map / Image generation / Editing and style transfer
Separating Style and Content for Generalized Style Transfer
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Increase training data from 20k to 300k images (ablation) | EMD · Chinese typeface transfer, D1 | L1 loss: 0.0096 -> 0.0091 | Helped |
| Increase training data from 20k to 300k images (ablation) | EMD · Chinese typeface transfer, D4 | L1 loss: 0.0095 -> 0.0094 | Helped |
| Increase training data from 300k to 500k images (ablation) | EMD · Chinese typeface transfer, D1 | L1 loss: 0.0091 -> 0.0091 | No clear effect |
| Increase reference set size from r=2 to r=10 (ablation) | EMD · Chinese typeface transfer, D1 | L1 loss: 0.0096 -> 0.0091 | Helped |
| Increase reference set size from r=10 to r=50 (ablation) | EMD · Chinese typeface transfer, D4 | L1 loss: 0.0094 -> 0.0093 | Helped |
| Add size/thickness and darkness weighting to the L1 loss (ablation) | EMD · Chinese typeface transfer, D4 | L1 loss: 0.0095 -> 0.0094 | Helped |
| Add content skip connections from encoder to decoder (ablation) | EMD · Chinese typeface transfer, novel contents | character detail and structure quality | Helped |
| Increase reference set size from r=10 to r=15 (ablation) | EMD · Chinese typeface transfer | generation quality | No clear effect |
| Use r=10 multi-image reference triplets instead of splitting them into r=1 triplets (ablation) | EMD · Chinese typeface transfer | generation quality | No clear effect |
| Use EMD rather than Pix2pix, AEGN, or CycleGAN | Chinese typeface transfer with known target styles and novel contents · Chinese typeface transfer | L1 loss | Helped |
| Use EMD with 10 style references rather than Zi-to-zi trained on 300 examples per style | Chinese typeface transfer with novel target styles · Chinese typeface transfer | L1 loss | Helped |
Closest papers by meaning
- Artistic Style Transfer with Internal-external Learning and Contrastive Learning (NeurIPS 2021 · similarity 0.90)
- Puff-Net: Efficient Style Transfer with Pure Content and Style Feature Fusion Network (CVPR 2024 · similarity 0.88)
- Style Injection in Diffusion: A Training-free Approach for Adapting Large-scale Diffusion Models for Style Transfer (CVPR 2024 · similarity 0.88)
- CSGO: Content-Style Composition in Text-to-Image Generation (NeurIPS 2025 · similarity 0.88)
- Z*: Zero-shot Style Transfer via Attention Reweighting (CVPR 2024 · similarity 0.88)
- SigStyle: Signature Style Transfer via Personalized Text-to-Image Models (AAAI 2025 · similarity 0.87)
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.