Research map / Image generation / Editing and style transfer
Decoder Network Over Lightweight Reconstructed Feature for Fast Semantic Style Transfer
Ming Lu, Hao Zhao, Anbang Yao, Feng Xu, Yurong Chen, Li Zhang
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 |
|---|---|---|---|
| replace backward optimization through the network with feature-space reconstruction plus decoder | semantic style transfer · Style transfer without segmentation masks (50 samples) | average time (s): 112s -> 1.35s | Improved |
| use feed-forward transfer net instead of the proposed method | style transfer · Style transfer without segmentation masks (50 samples) | average time (s): 1.35s -> 0.063s | Improved |
| replace Gatys2016 with the proposed method | portrait style transfer · Portrait user study | average time per image (s): 127.3s -> 1.48s | Improved |
| replace Gatys2016 with the proposed method | portrait style transfer · Portrait user study | average user score (1-5): 4.1 -> 3.8 | Mixed |
| replace Fast-Gatys2016 guided transfer net with the proposed method | portrait style transfer · Portrait user study | average user score (1-5): 1.5 -> 3.8 | Improved |
| add multi-layer style loss instead of single-layer style loss (ablation) | proposed method · Qualitative closeups (Fig. 10) | texture transfer | Improved |
| use relu4_1 instead of relu3_1 for decoder and fusion (ablation) | proposed method · Qualitative (Fig. 8) | color distribution | Hurt |
| use relu5_1 instead of relu3_1 (ablation) | proposed method · Qualitative (Fig. 12) | result quality | Hurt |
| fuse features by iterative reconstruction instead of neural patch swap (ablation) | proposed method · Qualitative (Fig. 11) | texture transfer | Improved |
| increase feature-fusion iterations N beyond 500 (ablation) | proposed method · Qualitative (Fig. 7) | convergence | No clear effect |
| add segmentation masks in feature space | proposed method · Qualitative | failure cases such as ground textures on sky | Improved |
Closest papers by meaning
- A Style-Aware Content Loss for Real-time HD Style Transfer (ECCV 2018 · similarity 0.89)
- Arbitrary Style Transfer With Style-Attentional Networks (CVPR 2019 · similarity 0.89)
- Rethinking and Improving the Robustness of Image Style Transfer (CVPR 2021 · similarity 0.89)
- Learning Linear Transformations for Fast Image and Video Style Transfer (CVPR 2019 · similarity 0.88)
- Puff-Net: Efficient Style Transfer with Pure Content and Style Feature Fusion Network (CVPR 2024 · similarity 0.88)
- SCSA: A Plug-and-Play Semantic Continuous-Sparse Attention for Arbitrary Semantic Style Transfer (CVPR 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.
