Research map / Label-efficient and robust learning / Semi-supervised learning
Scenimefy: Learning to Craft Anime Scene via Semi-Supervised Image-to-Image Translation
Yuxin Jiang, Liming Jiang, Shuai Yang, Chen Change Loy
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 |
|---|---|---|---|
| add semi-supervised pseudo-pair branch and StylePatchNCE to the unsupervised pipeline | Scenimefy · anime scene FID | FID: 61.973 (best baseline) -> 48.922 | Improved |
| freeze early StyleGAN layers during fine-tuning (ablation) | StyleGAN fine-tuning · pseudo pair semantic consistency | L_BCE: 4.45 -> 4.24 | Improved |
| add LPIPS perceptual loss to fine-tuning (ablation) | StyleGAN fine-tuning · pseudo pair semantic consistency | L_BCE: 4.53 -> 4.24 | Improved |
| add CLIP cosine-distance loss to fine-tuning (ablation) | StyleGAN fine-tuning · pseudo pair semantic consistency | L_BCE: 4.25 -> 4.24 | Improved |
| add patch-wise CLIP contrastive loss to fine-tuning (ablation) | StyleGAN fine-tuning · pseudo pair semantic consistency | L_BCE: 4.30 -> 4.24 | Improved |
| add unsupervised branch to the supervised branch (ablation) | Scenimefy I2I · semantic consistency | L_BCE: 4.76 -> 4.39 | Improved |
| add supervised pseudo-pair branch to the unsupervised branch (ablation) | Scenimefy I2I · semantic consistency | L_BCE: 4.50 -> 4.39 | Improved |
| add conditional discriminator D_P (ablation) | Scenimefy I2I · semantic consistency | L_BCE: 4.68 -> 4.39 | Improved |
| add StylePatchNCE loss to supervised branch (ablation) | Scenimefy I2I · semantic consistency | L_BCE: 4.72 -> 4.39 | Improved |
| add cosine decay of supervised loss weight (ablation) | Scenimefy I2I · semantic consistency | L_BCE: 4.58 -> 4.39 | Improved |
| use StylePatchNCE instead of L1 reconstruction loss in supervised branch (ablation) | Scenimefy · qualitative | visual quality | Improved |
Closest papers by meaning
- APISR: Anime Production Inspired Real-World Anime Super-Resolution (CVPR 2024 · similarity 0.87)
- StdGEN: Semantic-Decomposed 3D Character Generation from Single Images (CVPR 2025 · similarity 0.86)
- SSIMBaD: Sigma Scaling with SSIM-Guided Balanced Diffusion for AnimeFace Colorization (NeurIPS 2025 · similarity 0.86)
- Alchemist: Turning Public Text-to-Image Data into Generative Gold (NeurIPS 2025 · similarity 0.85)
- Advancing Manga Analysis: Comprehensive Segmentation Annotations for the Manga109 Dataset (CVPR 2025 · similarity 0.84)
- CSGO: Content-Style Composition in Text-to-Image Generation (NeurIPS 2025 · 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.
