# Text-to-Image Alignment on Decomposable-Captions-4k

Source: [Text-to-Image Alignment on Decomposable-Captions-4k](<https://www.sota2.com/research/sota/text-to-image-alignment-on-decomposable-captions-4k>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-02-26T05:46:42.868Z

Evaluation of Text-to-Image Alignment performance on the Decomposable-Captions-4k dataset, measured using human preference scores and alignment accuracy metrics.

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

| Scope | Recorded value |
| --- | --- |
| Task | Text-to-Image Alignment |
| Dataset (including any stated split) | Decomposable-Captions-4k |
| Metric | Human Preference Score |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 6 of 6 |
| Coverage | All results currently returned for this benchmark |

Evaluation settings may differ between results. The original papers and recorded configurations are needed to establish comparability. A shared metric name alone does not establish an identical protocol.

## Metrics

- [Human Preference Score](<https://www.sota2.com/research/sota/text-to-image-alignment-on-decomposable-captions-4k.md?metric=Human%20Preference%20Score>): Higher is better; unit: PERCENT.

- [Alignment Accuracy](<https://www.sota2.com/research/sota/text-to-image-alignment-on-decomposable-captions-4k.md?metric=Alignment%20Accuracy>): Higher is better; unit: PERCENT.

- [Pairwise Win Rate](<https://www.sota2.com/research/sota/text-to-image-alignment-on-decomposable-captions-4k.md?metric=Pairwise%20Win%20Rate>): Higher is better; unit: PERCENT.

- [Pairwise Tie Rate](<https://www.sota2.com/research/sota/text-to-image-alignment-on-decomposable-captions-4k.md?metric=Pairwise%20Tie%20Rate>): Higher is better; unit: PERCENT.

- [Pairwise Loss Rate](<https://www.sota2.com/research/sota/text-to-image-alignment-on-decomposable-captions-4k.md?metric=Pairwise%20Loss%20Rate>): Lower is better; unit: PERCENT.

## Recorded results

| Method | Human Preference Score | Result date | Report paper |
| --- | --- | --- | --- |
| Evaluate-and-Refine (PW + CA) | 90.25 | Not recorded | 68fa71611e209131cea50fa1 |
| Evaluate-and-Refine (PW) | 89.53 | Not recorded | 68fa71611e209131cea50fa1 |
| Attend-and-Excite | 85.94 | Not recorded | 68fa71611e209131cea50fa1 |
| Structure-Diffusion | 74.93 | Not recorded | 68fa71611e209131cea50fa1 |
| Stable-Diffusion | 72.98 | Not recorded | 68fa71611e209131cea50fa1 |
| Composable-Diffusion | 70.28 | Not recorded | 68fa71611e209131cea50fa1 |

### Evaluate-and-Refine (PW + CA)

Human Preference Score: 90.25. Result ID: 6991caee3a6c3148490a9fbd.

[Reporting paper](<https://www.sota2.com/research/paper/68fa71611e209131cea50fa1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa71611e209131cea50fa1.md>)

[Implementation](<https://1jsingh.github.io/divide-evaluate-and-refine>)

Recorded evaluation configuration: {"Refinement":"Prompt-weighting and Cross-attention updates","Inference Time (s)":"12.24 s"}. This record does not assert independent reproduction.

### Evaluate-and-Refine (PW)

Human Preference Score: 89.53. Result ID: 6991caee3a6c3148490a9fbc.

[Reporting paper](<https://www.sota2.com/research/paper/68fa71611e209131cea50fa1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa71611e209131cea50fa1.md>)

[Implementation](<https://1jsingh.github.io/divide-evaluate-and-refine>)

Recorded evaluation configuration: {"Refinement":"Prompt-weighting","Inference Time (s)":"10.32 s"}. This record does not assert independent reproduction.

### Attend-and-Excite

Human Preference Score: 85.94. Result ID: 6991caee3a6c3148490a9fbb.

[Reporting paper](<https://www.sota2.com/research/paper/68fa71611e209131cea50fa1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6f961e209131cea46798.md>)

Recorded evaluation configuration: {"Inference Time (s)":"8.59 s"}. This record does not assert independent reproduction.

### Structure-Diffusion

Human Preference Score: 74.93. Result ID: 6991caee3a6c3148490a9fba.

[Reporting paper](<https://www.sota2.com/research/paper/68fa71611e209131cea50fa1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6ea71e209131cea411f7.md>)

Recorded evaluation configuration: {"Inference Time (s)":"11.51 s"}. This record does not assert independent reproduction.

### Stable-Diffusion

Human Preference Score: 72.98. Result ID: 6991caee3a6c3148490a9fb8.

[Reporting paper](<https://www.sota2.com/research/paper/68fa71611e209131cea50fa1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6c391e209131cea32e4a.md>)

[Implementation](<https://github.com/CompVis/latent-diffusion>)

Recorded evaluation configuration: {"Inference Time (s)":"3.54 s"}. This record does not assert independent reproduction.

### Composable-Diffusion

Human Preference Score: 70.28. Result ID: 6991caee3a6c3148490a9fb9.

[Reporting paper](<https://www.sota2.com/research/paper/68fa71611e209131cea50fa1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6e5a1e209131cea3f4ee.md>)

Recorded evaluation configuration: {"Inference Time (s)":"10.89 s"}. This record does not assert independent reproduction.

Page 1 of 1. This is an index page, not the full collection.
