Research map / Image generation / Text-to-image diffusion
Consis-GCPO: Consistency-Preserving Group Causal Preference Optimization for Vision Customization
Qiaoqiao Jin, Dong She, Siming Fu, Hualiang Wang, Mu Liu, Jidong Jiang
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 full decoupled causal interventions to Flow-GRPO (ablation) | Flow-GRPO · Multi-subject R2I | DINO-I: 0.551 -> 0.572 | Improved |
| add full decoupled causal interventions to Flow-GRPO (ablation) | Flow-GRPO · Multi-subject R2V | CLIP-I: 0.645 -> 0.674 | Improved |
| add prompt-only interventions to Flow-GRPO (ablation) | Flow-GRPO · Multi-subject R2I | CLIP-T: 0.325 -> 0.338 | Improved |
| add prompt-only interventions to Flow-GRPO (ablation) | Flow-GRPO · Multi-subject R2I | DINO-I: 0.551 -> 0.544 | Hurt |
| add reference-only interventions to Flow-GRPO (ablation) | Flow-GRPO · Multi-subject R2I | CLIP-I: 0.742 -> 0.780 | Improved |
| add reference-only interventions to Flow-GRPO (ablation) | Flow-GRPO · Multi-subject R2V | CLIP-T: 0.265 -> 0.255 | Hurt |
| use joint instead of alternating optimization of text and image rewards (ablation) | Consis-GCPO · DreamBench (single-subject) | DINO-I: 0.762 -> 0.781 | Improved |
| use joint instead of sequential optimization (ablation) | Consis-GCPO · DreamBench (single-subject) | CLIP-T: 0.308 -> 0.325 | Improved |
| joint optimization shares backward pass (ablation) | Consis-GCPO · DreamBench (single-subject) | training cost: 1.8x slower (alternating) -> 1.0x | Improved |
| raise lambda_Ir from 1.0 to 2.0 (ablation) | Consis-GCPO · DreamBench (single-subject) | CLIP-T: 0.325 -> 0.305 | Hurt |
| use softmax temperature 1.0 instead of 0.8 (ablation) | Consis-GCPO · DreamBench (single-subject) | DINO-I: 0.765 -> 0.781 | Improved |
| use softmax temperature 1.0 instead of 1.2 (ablation) | Consis-GCPO · DreamBench (single-subject) | DINO-I: 0.772 -> 0.781 | Improved |
Closest papers by meaning
- Reinforcement Learning Meets Masked Generative Models: Mask-GRPO for Text-to-Image Generation (NeurIPS 2025 · similarity 0.87)
- Seeing What Matters: Visual Preference Policy Optimization for Visual Generation (CVPR 2026 · similarity 0.87)
- Learning What to Trust: Bayesian Prior-Guided Optimization for Visual Generation (CVPR 2026 · similarity 0.87)
- Multimodal Large Language Models Make Text-to-Image Generative Models Align Better (NeurIPS 2024 · similarity 0.86)
- Dynamics-Aware Preference Optimization for Vision-Language Models (CVPR 2026 · similarity 0.86)
- MaskSense: Confronting the Visual Exploration Trap in Masked Image Generation (NeurIPS 2026 · similarity 0.86)
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.
