Research map / Image generation / Text-to-image diffusion
Discovering Divergent Representations between Text-to-Image Models
Lisa Dunlap, Joseph E. Gonzalez, Trevor Darrell, Fabian Caba Heilbron, Josef Sivic, Bryan Russell
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 |
|---|---|---|---|
| run five evolutionary iterations instead of one for description discovery (ablation) | CompCon · ID2 (Top 1) | description score: 0.59 -> 0.64 | Improved |
| run five iterations instead of one (Top 5) (ablation) | CompCon · ID2 (Top 5) | description score: 0.72 -> 0.78 | Improved |
| CompCon attribute discovery instead of VisDiff | attribute discovery · ID2 (Top 1) | attribute score: 0.47 -> 0.60 | Improved |
| CompCon attribute discovery instead of TF-IDF | attribute discovery · ID2 (Top 1) | attribute score: 0.23 -> 0.60 | Improved |
| CompCon instead of LLM-only end-to-end | CompCon · ID2 (Top 1) | description score: 0.03 -> 0.64 | Improved |
| use IDEFICS open VLM instead of GPT-4o VLM (ablation) | CompCon · ID2 (Top 1) | attribute score: 0.60 -> 0.56 | Hurt |
| use IDEFICS plus llama3-8b instead of GPT-4o for both (ablation) | CompCon · ID2 (Top 1) | attribute score: 0.60 -> 0.39 | Hurt |
| use IDEFICS plus llama3-8b instead of GPT-4o for both (ablation) | CompCon · ID2 (Top 1) | description score: 0.64 -> 0.46 | Hurt |
| inject 25% VLM errors in attribute discovery (ablation) | CompCon · ID2 (Top 1) | attribute score: 0.60 -> 0.56 | No clear effect |
| more evolutionary iterations raise proportion of diverging prompts (Mandala) (ablation) | CompCon · Mandala diverging prompts | proportion diverging: 12% -> 76% | Improved |
| more iterations (Wet Streets) (ablation) | CompCon · Wet Streets diverging prompts | proportion diverging: 8% -> 52% | Improved |
| more iterations (Decay) (ablation) | CompCon · Decay diverging prompts | proportion diverging: 8% -> 44% | Improved |
Closest papers by meaning
- Localizing and Editing Knowledge In Text-to-Image Generative Models (ICLR 2024 · similarity 0.87)
- ConceptBed: Evaluating Concept Learning Abilities of Text-to-Image Diffusion Models (AAAI 2024 · similarity 0.87)
- The Hidden Language of Diffusion Models (ICLR 2024 · similarity 0.86)
- The Many Senses of Visual Similarity: A Text-Prompted Image Perceptual Metric (NeurIPS 2026 · similarity 0.86)
- Attribute Based Interpretable Evaluation Metrics for Generative Models (ICML 2024 · similarity 0.86)
- Magnet: We Never Know How Text-to-Image Diffusion Models Work, Until We Learn How Vision-Language Models Function (NeurIPS 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.
