Research map / Image generation / Text-to-image diffusion
PureCC: Pure Learning for Text-to-Image Concept Customization
Zhichao Liao, Xiaole Xian, Qingyu Li, Wenyu Qin, Meng Wang, Weicheng Xie, Siyang Song, Pingfa Feng, Long Zeng, Liang Pan
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 L_PureCC loss to L_CC (ablation) | PureCC · DreamBenchPCC (instance) | Seg-Cons: 23.74 -> 69.37 | Improved |
| add L_PureCC loss to L_CC (ablation) | PureCC · DreamBenchPCC (instance) | dCLIP-T (base): -4.52 -> -0.31 | Improved |
| add L_PureCC loss to L_CC (ablation) | PureCC · DreamBenchPCC (instance) | CLIP-I (target): 0.65 -> 0.81 | Improved |
| add separate pre-fine-tuned frozen extractor (two-stage) instead of merged training (ablation) | PureCC · DreamBenchPCC (instance) | CLIP-I (target): 0.50 -> 0.81 | Improved |
| add separate pre-fine-tuned frozen extractor (two-stage) instead of merged training (ablation) | PureCC · DreamBenchPCC (instance) | Seg-Cons: 54.37 -> 69.37 | Hurt |
| use adaptive lambda* instead of fixed lambda=1.0 (ablation) | PureCC · DreamBenchPCC (instance) | CLIP-I (target): 0.43 -> 0.81 | Improved |
| use adaptive lambda* instead of fixed lambda=5.0 (ablation) | PureCC · DreamBenchPCC (instance) | dCLIP-T (base): -2.67 -> -0.31 | Improved |
| use adaptive lambda* instead of fixed lambda=3.0 (ablation) | PureCC · DreamBenchPCC (style) | CSD: 0.61 -> 0.63 | Improved |
| raise eta from 0.5 to 1.0 (ablation) | PureCC · DreamBenchPCC (instance) | CLIP-I (target): 0.55 -> 0.81 | Improved |
| raise eta from 1.0 to 2.0 (ablation) | PureCC · DreamBenchPCC (instance) | CLIP-I (target): 0.81 -> 0.67 | Hurt |
| raise eta from 1.0 to 1.5 (ablation) | PureCC · DreamBenchPCC (instance) | CLIP-I (target): 0.81 -> 0.80 | No clear effect |
| use trainable model's own base-text prediction as original prediction instead of extra frozen model (ablation) | PureCC · DreamBenchPCC (qualitative) | visual quality | No clear effect |
Closest papers by meaning
- Concept Conductor: Orchestrating Multiple Personalized Concepts in Text-to-Image Synthesis (AAAI 2025 · similarity 0.86)
- MC^2: Multi-concept Guidance for Customized Multi-concept Generation (CVPR 2025 · similarity 0.85)
- How to Continually Adapt Text-to-Image Diffusion Models for Flexible Customization? (NeurIPS 2024 · similarity 0.85)
- ConceptGuard: Continual Personalized Text-to-Image Generation with Forgetting and Confusion Mitigation (CVPR 2025 · similarity 0.84)
- Concept Weaver: Enabling Multi-Concept Fusion in Text-to-Image Models (CVPR 2024 · similarity 0.84)
- ClassDiffusion: More Aligned Personalization Tuning with Explicit Class Guidance (ICLR 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.
