Research map / Image generation / Text-to-image diffusion
Region-Level Data Attribution for Text-to-Image Generative Models
Trong Bang Nguyen, Phi Le Nguyen, Simon Lucey, Minh Hoai
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 the second training phase with positive and negative samples (ablation) | AR detector (B-B) · Backbone ablation | Recall@5: 0.4425 -> 0.4696 | Improved |
| add the second training phase (ablation) | AR detector (T-B) · Backbone ablation | Recall@5: 0.2701 -> 0.2910 | Improved |
| add the second training phase (ablation) | AR detector (B-L) · Backbone ablation | Recall@5: 0.3124 -> 0.3231 | Improved |
| add the second training phase (ablation) | AR detector (B-B) · Backbone ablation | Recall@10: 0.704 -> 0.709 | Improved |
| use SwinT-Tiny image backbone instead of SwinT-Base (ablation) | AR detector · Backbone ablation | Recall@5: 0.4696 -> 0.2910 | Hurt |
| use BERT-Large text backbone instead of BERT-Base (ablation) | AR detector · Backbone ablation | Recall@5: 0.4696 -> 0.3231 | Hurt |
| region-level AR detector instead of image-level AbC | attribution · COCO | Recall@10: 0.2079 -> 0.7278 | Improved |
| region-level AR detector instead of image-level AbC | attribution · LVIS | Recall@10: 0.1061 -> 0.3516 | Improved |
| AR detector instead of few-shot detector ICPE | attribution · COCO | Box acc.: 0.86 -> 0.89 | Improved |
| AR detector instead of few-shot detector ICPE | attribution · LVIS | Box acc.: 0.62 -> 0.72 | Improved |
Closest papers by meaning
- Data Attribution for Text-to-Image Models by Unlearning Synthesized Images (NeurIPS 2024 · similarity 0.90)
- Fast Data Attribution for Text-to-Image Models (NeurIPS 2025 · similarity 0.89)
- Less is More: Fewer Interpretable Region via Submodular Subset Selection (ICLR 2024 · similarity 0.87)
- EvolveDirector: Approaching Advanced Text-to-Image Generation with Large Vision-Language Models (NeurIPS 2024 · similarity 0.86)
- quanda: An Interpretability Toolkit for Training Data Attribution Evaluation (NeurIPS 2026 · similarity 0.86)
- Localizing and Editing Knowledge In Text-to-Image Generative Models (ICLR 2024 · similarity 0.85)
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.
