Research map / Vision-language and multimodal / Human and embodied modeling
Visual Personalization Turing Test
Rameen Abdal, James Burgess, Sergey Tulyakov, Kuan-Chieh Jackson Wang
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 |
|---|---|---|---|
| replace BRAG with VPRAG hierarchical retrieval | persona prompt rewriting · VPTT human study (6000 ratings, generation and editing) | Human winning accuracy: 21.3% -> 62.0% | Improved |
| replace BRAG with VPRAG | persona prompt rewriting · VPTT human study (6000 ratings, generation and editing) | VPTT_score-c average: 0.420 -> 0.464 | Improved |
| replace BRAG with VPRAG | persona prompt rewriting · VPTT human study (6000 ratings, generation and editing) | Human rating average: 2.69 -> 3.34 | Improved |
| add persona demographics to the prompt (Persona Only vs Baseline) | persona prompt rewriting · VPTT human study (6000 ratings, generation and editing) | Human rating average: 1.64 -> 2.51 | Improved |
| replace BRAG with VPRAG | Qwen2.5-7B-Instruct · VPTT-Bench 10,000 personas (generation) | VPTT_score: 0.581 -> 0.631 | Improved |
| combine BRAG with VPRAG instead of VPRAG alone | GPT-4o-mini · VPTT-Bench 10,000 personas (generation) | VPTT_score: 0.640 -> 0.644 | No clear effect |
| combine BRAG with VPRAG instead of VPRAG alone | Qwen2.5-7B-Instruct · VPTT-Bench 10,000 personas (generation) | VPTT_score: 0.631 -> 0.602 | Hurt |
| replace BRAG with VPRAG | GPT-4o-mini · VPTT-Bench 10,000 personas (editing) | VPTT_score: 0.596 -> 0.626 | Improved |
| combine BRAG with VPRAG instead of VPRAG alone | 200 personas (generation) · 200 personas, 3 LLM rewriters (generation) | VLM score: 4.08 -> 4.30 | Improved |
| combine BRAG with VPRAG instead of VPRAG alone | 200 personas (editing) · 200 personas, 3 LLM rewriters (editing) | VPTT_score-c winning accuracy: 47.2% -> 28.3% | Hurt |
| use text-only VPTT_score-c instead of VLM judge as proxy for humans | metric validity · human study | Spearman correlation with human (combined): 0.67 -> 0.68 | No clear effect |
| evaluate editing instead of generation | VPTT_score-c · human study | Spearman correlation with human: 0.78 -> 0.5 | Hurt |
Closest papers by meaning
- MMPB: It’s Time for Multi-Modal Personalization (NeurIPS 2025 · similarity 0.86)
- Contextualized Visual Personalization in Vision-Language Models (ICML 2026 · similarity 0.85)
- DreamBench++: A Human-Aligned Benchmark for Personalized Image Generation (ICLR 2025 · similarity 0.84)
- Personalized Representation from Personalized Generation (ICLR 2025 · similarity 0.83)
- Personal Visual Memory from Explicit and Implicit Evidence (NeurIPS 2026 · similarity 0.83)
- MaSC: A Masked Similarity Metric for Evaluating Concept-Driven Generation (NeurIPS 2026 · similarity 0.83)
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.
