AI Paper Atlas

Research map / Image generation / Text-to-image diffusion

Divide, Evaluate, and Refine: Evaluating and Improving Text-to-Image Alignment with Iterative VQA Feedback

Jaskirat Singh, Liang Zheng

NeurIPS 2023 · Image generation › Text-to-image diffusion · approach: Inference-time T2I improvement · 2 citations (OpenAlex)

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 changedWhereMeasuredEffect
add cross-attention updates on top of prompt weighting (ablation)Ours (PW) · Human evaluationalignment accuracy (%): 70.28 -> 74.16Improved
add cross-attention updates on top of prompt weighting (ablation)Ours (PW) · Human evaluationnormalized human score (%): 89.53 -> 90.25Improved
add cross-attention updates on top of prompt weighting (ablation)Ours (PW) · Human evaluationinference time per image (s): 10.32 s -> 12.24 sHurt
iterative prompt-weighting refinement vs Attend-and-ExciteOurs (PW) · Human evaluationalignment accuracy (%): 65.50 -> 70.28Improved
full method (PW + CA) vs Attend-and-ExciteStable Diffusion v1.4 · Human evaluationalignment accuracy (%): 65.50 -> 74.16Improved
iterative refinement vs plain Stable DiffusionStable Diffusion v1.4 · Human evaluationalignment accuracy (%): 43.66 -> 74.16Improved
iterative refinement vs plain Stable DiffusionStable Diffusion v1.4 · Human evaluationinference time per image (s): 3.54 s -> 12.24 sHurt
full method vs Attend-and-Excite on very hard promptsStable Diffusion v1.4 · Very hard promptsalignment accuracy (%): 49.5 -> 62.9Improved
increase prompt difficultyall methods · Prompt difficulty levelsalignment accuracy (%)Hurt
Composable-Diffusion vs Stable DiffusionStable Diffusion v1.4 · Human evaluationalignment accuracy (%): 43.66 -> 37.72Hurt
Structure-Diffusion vs Stable DiffusionStable Diffusion v1.4 · Human evaluationalignment accuracy (%): 43.66 -> 45.23No clear effect
HPS human-feedback finetuned Stable Diffusion vs originalStable Diffusion v1.4 · Prompt alignmentsemantic alignmentNo clear effect

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy