Research map / Image generation / Editing and style transfer
Cognitive Enhancement Chain-of-Thought Towards Enhancing Style Learning and Content Preservation for Long Style Transfer
Lianwei Wu, Botao Wang, Wenbo An, Tieqiao Li, Xianghua Li
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 LLM-driven source content preserving (progressive CoT) (ablation) | CeCoT · ER->SP (English) | Overall2: 60.0 -> 72.6 | Improved |
| add LLM-driven target style learning (cognitive CoT) (ablation) | CeCoT · ER->SP (English) | Overall2: 48.7 -> 72.6 | Improved |
| add fusion network (ablation) | CeCoT · ER->SP (English) | Overall2: 69.8 -> 72.6 | Improved |
| add content preservation enhancing loss (ablation) | CeCoT · ER->SP (English) | Overall2: 68.4 -> 72.6 | Improved |
| add style consistency verification loss (ablation) | CeCoT · ER->SP (English) | Overall2: 53.3 -> 72.6 | Improved |
| add style consistency verification loss (ablation) | CeCoT · ER->SP (English) | document-level style accuracy: 35.6 -> 62.1 | Improved |
| use designed CoT prompts instead of standard prompts in TSL and SCP (ablation) | CeCoT · Chinese/English datasets (Figure 4) | Overall metrics | Improved |
| remove progressive CoT prompts P'1 or P'2 from source content preserving (ablation) | CeCoT · Figure 4 | all metrics | Hurt |
| remove any one prompt-order mode in target style learning (ablation) | CeCoT · Figure 4 | Overall | Hurt |
| use hybrid mode 5 vs serial mode 1 ordering (ablation) | CeCoT · Figure 5 (Overall2 by sample size) | Overall2 | Mixed |
| use CeCoT instead of SC2 | long text style transfer · FT->LX (Chinese) | Overall2: 51.4 -> 53.4 | Improved |
| use CeCoT instead of SC2 | long text style transfer · ER->SP (English) | Overall1: 25.3 -> 30.9 | Improved |
Closest papers by meaning
- Style Transfer in Text: Exploration and Evaluation (AAAI 2018 · similarity 0.88)
- UniCSG: Unified High-Fidelity content-constrained style-driven generation via Staged Semantic and Frequency Disentanglement (ECCV 2026 · similarity 0.87)
- ParaGuide: Guided Diffusion Paraphrasers for Plug-and-Play Textual Style Transfer (AAAI 2024 · similarity 0.85)
- Adaptive Prompt Routing for Arbitrary Text Style Transfer with Pre-trained Language Models (AAAI 2024 · similarity 0.84)
- Retaining Knowledge and Enhancing Long-Text Representations in CLIP through Dual-Teacher Distillation (CVPR 2025 · similarity 0.83)
- K-LoRA: Unlocking Training-Free Fusion of Any Subject and Style LoRAs (CVPR 2025 · 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.
