Research map / Image generation / Text-to-image diffusion
AITTI: Learning Adaptive Inclusive Token for Text-to-Image Generation
Xinyu Hou, Xiaoming Li, Chen Change Loy
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 anchor loss (no mapping network) (ablation) | rTI · Race bias | D_KL: 0.8230 -> 0.3461 | Improved |
| add anchor loss (no mapping network) (ablation) | rTI · Gender bias | D_KL: 0.3124 -> 0.3121 | No clear effect |
| add adaptive mapping network on top of anchor loss (ablation) | anchor-loss-only · Gender bias | D_KL: 0.3121 -> 0.1290 | Improved |
| add adaptive mapping network on top of anchor loss (ablation) | anchor-loss-only · Race bias | D_KL: 0.3461 -> 0.3488 | No clear effect |
| use Transformer instead of MLP mapping network (ablation) | AITTI · Gender bias | D_KL: 0.4059 -> 0.1290 | Improved |
| place mapping module before text transformer instead of after (ablation) | AITTI · Gender bias | D_KL: 0.3388 -> 0.1290 | Improved |
| use 1 inclusive token instead of 2 (ablation) | AITTI · Gender bias | D_KL: 0.2074 -> 0.1290 | Improved |
| use 10 instead of 5 training samples per combination (ablation) | AITTI · Gender bias | D_KL: 0.1988 -> 0.1290 | Improved |
| use 10 instead of 20 training samples per combination (ablation) | AITTI · Gender bias | D_KL: 0.1573 -> 0.1290 | Mixed |
| train on SD1.5 synthetic images instead of online images (ablation) | AITTI · Gender bias | D_KL: 0.2386 -> 0.1520 | Improved |
| use 4 transformer blocks instead of 2 (ablation) | adaptive mapping network · Gender bias | D_KL: 0.1630 -> 0.1290 | Mixed |
| combine gender, race and age inclusive tokens | AITTI · Multi-bias | D_KL: 0.2319 -> 0.2790 | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.
- ITI-GEN — ITI-GEN: Inclusive Text-to-Image Generation (ICCV 2023) · better on 1 of 2 shared results
- TIME — TIME: Text and Image Mutual-Translation Adversarial Networks (AAAI 2021) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Rethinking Training for De-biasing Text-to-Image Generation: Unlocking the Potential of Stable Diffusion (CVPR 2025 · similarity 0.88)
- FairImagen: Post-Processing for Bias Mitigation in Text-to-Image Models (NeurIPS 2025 · similarity 0.87)
- Fair Text-to-Image Diffusion via Fair Mapping (AAAI 2025 · similarity 0.87)
- Boosting Alignment for Post-Unlearning Text-to-Image Generative Models (NeurIPS 2024 · similarity 0.86)
- OneIG-Bench: Omni-dimensional Nuanced Evaluation for Image Generation (NeurIPS 2025 · similarity 0.86)
- OpenBias: Open-set Bias Detection in Text-to-Image Generative Models (CVPR 2024 · 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.
