Research map / Image generation / Text-to-image diffusion
Enhancing prompt-image alignment evaluations via cyclic mutual information maximization
Xingran Liao, Duanyu Feng, Mingliang Zhou, Sam Kwong, Weisi Lin
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 forward and backward MIM over the fusion baseline (ablation) | CMIM-AIGIQA · AGIQA-3K alignment | SRCC: 0.7731 -> 0.8546 | Improved |
| add forward MIM (alignment) (ablation) | CMIM-AIGIQA · AGIQA-3K alignment | SRCC: 0.8415 -> 0.8546 | Improved |
| add forward MIM (quality) (ablation) | CMIM-AIGIQA · AGIQA-3K quality | SRCC: 0.8432 -> 0.8755 | Improved |
| add backward MIM (ablation) | CMIM-AIGIQA · AIGCIQA2023 alignment | SRCC: 0.7655 -> 0.8263 | Improved |
| add textual entropy H(Y) (ablation) | CMIM-AIGIQA · AGIQA-3K alignment | SRCC: 0.8502 -> 0.8546 | Improved |
| add textual entropy H(Y) (ablation) | CMIM-AIGIQA · PKU-AIGIQA-4k alignment | SRCC: 0.8094 -> 0.8223 | Improved |
| use NCE loss instead of MSE in backward phase (ablation) | CMIM-AIGIQA · AGIQA-3K alignment | SRCC: 0.8156 -> 0.8546 | Improved |
| use NCE loss instead of MSE in backward phase (ablation) | CMIM-AIGIQA · AIGCIQA2023 alignment | SRCC: 0.7892 -> 0.8263 | Improved |
| add cross-attention (CSA) fusion (ablation) | CMIM-AIGIQA · PKU-AIGIQA-4k alignment | SRCC: 0.7984 -> 0.8223 | Improved |
| use BLIP backbone instead of CLIP ViT-L/14 (alignment) (ablation) | CMIM-AIGIQA · AGIQA-3K alignment | SRCC: 0.8214 -> 0.8546 | Improved |
| use BLIP backbone instead of CLIP ViT-L/14 (quality) (ablation) | CMIM-AIGIQA · AGIQA-3K quality | SRCC: 0.8931 -> 0.8755 | Hurt |
| use CMIM-AIGIQA instead of MoE-AGIQA in cross-dataset transfer | CMIM-AIGIQA vs MoE-AGIQA · PKU-AIGIQA-4k to AIGCIQA2023 | SRCC: 0.5646 -> 0.6305 | Improved |
Closest papers by meaning
- Information Theoretic Text-to-Image Alignment (ICLR 2025 · similarity 0.88)
- Amend to Alignment: Decoupled Prompt Tuning for Mitigating Spurious Correlation in Vision-Language Models (ICML 2024 · similarity 0.88)
- PromptCoT: Align Prompt Distribution via Adapted Chain-of-Thought (CVPR 2024 · similarity 0.87)
- Seeing the Image: Prioritizing Visual Correlation by Contrastive Alignment (NeurIPS 2024 · similarity 0.87)
- Noise Diffusion for Enhancing Semantic Faithfulness in Text-to-Image Synthesis (CVPR 2025 · similarity 0.86)
- Salient Concept-Aware Generative Data Augmentation (NeurIPS 2025 · 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.
