Research map / Image generation / Editing and style transfer
Learning to Incorporate Texture Saliency Adaptive Attention to Image Cartoonization
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 |
|---|---|---|---|
| add patch-level adversarial branch with CTSS-sampled texture-salient patches on top of image-level adversarial loss (ablation) | image cartoonization GAN · TWR / DB / CSC | FID: 145.25 -> 115.25 (TWR); 146.63 -> 112.97 (DB) in ablation vs full | Helped |
| remove image-level adversarial branch (patch-level only) (ablation) | image cartoonization GAN · TWR | FID: 182.78 vs 115.25 | Hurt |
| add CTSS saliency sampling instead of using all local patches (ablation) | patch-level adversarial branch · TWR / DB / CSC | FID: 127.09 -> 115.25 (TWR); 125.80 -> 112.97 (DB); 130.90 -> 124.66 (CSC) | Helped |
| edge-promoting adversarial loss (CartoonGAN) in place of basic global adversarial loss, with no local branch (ablation) | image-level adversarial loss · TWR | FID: 145.25 -> 143.40 | Helped |
| add Gram style loss to global adversarial loss (ablation) | image-level adversarial loss · TWR | FID: 145.25 -> 139.05 | Helped |
| add Gram and mean-variance style losses to edge-promoting loss (ablation) | image-level adversarial loss · TWR | FID: 143.40 -> 135.45 | Helped |
| increase number of sampled patches K from 32 to 128 (ablation) | CTSS patch sampler · TWR / DB / CSC | FID: 115.25 -> 126.78 (TWR); 112.97 -> 125.64 (DB) | Hurt |
| decrease number of sampled patches K from 32 to 16 (ablation) | CTSS patch sampler · TWR / DB / CSC | FID: 115.25 -> 115.86 (TWR); 112.97 -> 113.53 (DB); 124.66 -> 125.45 (CSC) | Hurt |
| full method (CTSS patch-level branch + image-level branch) vs prior cartoonization methods | image cartoonization · TWR / DB / CSC | FID: best baseline WhiteBox 132.67 / AnimeGAN 134.94 / AnimeGAN 138.83 -> Ours 115.25 / 112.97 / 124.66 | Helped |
Closest papers by meaning
- CartoonGAN: Generative Adversarial Networks for Photo Cartoonization (CVPR 2018 · similarity 0.92)
- Learning to Cartoonize Using White-Box Cartoon Representations (CVPR 2020 · similarity 0.89)
- Interactive Cartoonization With Controllable Perceptual Factors (CVPR 2023 · similarity 0.88)
- ImageNet-trained CNNs are not biased towards texture: Revisiting feature reliance through controlled suppression (NeurIPS 2025 · similarity 0.85)
- Attention Distillation: A Unified Approach to Visual Characteristics Transfer (CVPR 2025 · similarity 0.84)
- Free-Lunch Color-Texture Disentanglement for Stylized 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.