Research map / Image generation / Editing and style transfer
AesFormer: Transform Everyday Photos into Beautiful Memories
Tianxiang Du, Hulingxiao He, Yuxin Peng
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 SFT-trained AesThinker action planner (S1a) (ablation) | Qwen-Image-Edit-2511 · AesRecon, win rate vs poor (GPT-4o) | win rate: 16.50 -> 61.04 | Improved |
| add GRPO-A stage (S1b) to SFT thinker and editor (ablation) | AesFormer · AesRecon, win rate vs poor (GPT-4o) | win rate: 61.13 -> 65.33 | Improved |
| add GRPO-A stage (ablation) | AesFormer · AesRecon, win rate vs good (GPT-4o) | win rate: 24.14 -> 26.25 | Improved |
| add Stage 2 AesEditor fine-tuning to SFT thinker (ablation) | AesThinker S1a · AesRecon, win rate vs poor (GPT-4o) | win rate: 61.04 -> 61.13 | No clear effect |
| use ordered seven-dimension actions instead of shuffled order (ablation) | AesThinker S1a · AesRecon, win rate vs poor (GPT-4o) | win rate: 58.69 -> 61.04 | Improved |
| use ordered seven-dimension actions instead of shuffled order (ablation) | AesThinker S1a · AesRecon, win rate vs good (GPT-4o) | win rate: 18.60 -> 24.58 | Improved |
| use ordered seven-dimension actions instead of shuffled order (ablation) | AesThinker S1a · AesRecon, ArtiMuse | ArtiMuse score: 46.16 -> 47.70 | Improved |
| add off-the-shelf GPT-4o thinker | Qwen-Image-Edit-2511 · AesRecon, win rate vs poor (GPT-4o) | win rate: 16.50 -> 14.62 | Hurt |
| add off-the-shelf Qwen3 thinker | Bagel · AesRecon, win rate vs poor (GPT-4o) | win rate: 12.40 -> 3.43 | Hurt |
| add off-the-shelf GPT-4o thinker | Step1X-Edit-v1.1 · AesRecon, win rate vs poor (GPT-4o) | win rate: 15.28 -> 6.20 | Hurt |
| add off-the-shelf GPT-4o thinker | FLUX.1 Kontext [Dev] · AesRecon, win rate vs poor (human) | win rate: 5.88 -> 13.72 | Improved |
| add off-the-shelf GPT-4o thinker | FLUX.1 Kontext [Dev] · AesRecon, win rate vs poor (GPT-4o) | win rate: 12.96 -> 12.51 | No clear effect |
Closest papers by meaning
- Venus: Benchmarking and Empowering Multimodal Large Language Models for Aesthetic Guidance and Cropping (CVPR 2026 · similarity 0.87)
- ArtEditor: Learning Customized Instructional Image Editor from Few-Shot Examples (ICCV 2025 · similarity 0.86)
- PhotoFramer: Multi-modal Image Composition Instruction (CVPR 2026 · similarity 0.86)
- AccelAes: Accelerating Diffusion Transformers for Training-Free Aesthetic-Enhanced Image Generation (ECCV 2026 · similarity 0.86)
- AesFA: An Aesthetic Feature-Aware Arbitrary Neural Style Transfer (AAAI 2024 · similarity 0.85)
- JarvisArt: Liberating Human Artistic Creativity via an Intelligent Photo Retouching Agent (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.
