# Text-to-Image Preference Evaluation on HPD v2 (test)

Source: [Text-to-Image Preference Evaluation on HPD v2 (test)](<https://www.sota2.com/research/sota/text-to-image-preference-evaluation-on-hpd-v2-test>)

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

Data updated: 2026-02-26T05:34:43.491Z

Evaluation of Text-to-Image model outputs based on human preference scores derived from the HPD v2 test set, utilizing metrics such as Aesthetic, PickScore, HPS, and CLIP scores.

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 Preference Evaluation |
| Dataset (including any stated split) | HPD v2 (test) |
| Metric | Aesthetic (Mean) |
| Unit | SCALAR |
| Direction | Higher is better |
| Results loaded | 14 of 14 |
| 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

- [Aesthetic (Mean)](<https://www.sota2.com/research/sota/text-to-image-preference-evaluation-on-hpd-v2-test.md?metric=Aesthetic%20(Mean)>): Higher is better; unit: SCALAR.

- [Aesthetic (Median)](<https://www.sota2.com/research/sota/text-to-image-preference-evaluation-on-hpd-v2-test.md?metric=Aesthetic%20(Median)>): Higher is better; unit: SCALAR.

- [PickScore (Median)](<https://www.sota2.com/research/sota/text-to-image-preference-evaluation-on-hpd-v2-test.md?metric=PickScore%20(Median)>): Higher is better; unit: SCALAR.

- [PickScore (Mean)](<https://www.sota2.com/research/sota/text-to-image-preference-evaluation-on-hpd-v2-test.md?metric=PickScore%20(Mean)>): Higher is better; unit: SCALAR.

- [HPS (Median)](<https://www.sota2.com/research/sota/text-to-image-preference-evaluation-on-hpd-v2-test.md?metric=HPS%20(Median)>): Higher is better; unit: SCALAR.

- [HPS (Mean)](<https://www.sota2.com/research/sota/text-to-image-preference-evaluation-on-hpd-v2-test.md?metric=HPS%20(Mean)>): Higher is better; unit: SCALAR.

- [CLIP Score (Median)](<https://www.sota2.com/research/sota/text-to-image-preference-evaluation-on-hpd-v2-test.md?metric=CLIP%20Score%20(Median)>): Higher is better; unit: SCALAR.

- [CLIP Score (Mean)](<https://www.sota2.com/research/sota/text-to-image-preference-evaluation-on-hpd-v2-test.md?metric=CLIP%20Score%20(Mean)>): Higher is better; unit: SCALAR.

- [HPSv2 (Mean)](<https://www.sota2.com/research/sota/text-to-image-preference-evaluation-on-hpd-v2-test.md?metric=HPSv2%20(Mean)>): Higher is better; unit: SCALAR.

## Recorded results

| Method | Aesthetic (Mean) | Result date | Report paper |
| --- | --- | --- | --- |
| InPO-SDXL | 6.182 | Not recorded | 68fa77911e209131cea73f9f |
| Base-SDXL | 6.1346 | Not recorded | 68fa77911e209131cea73f9f |
| DPO-SDXL | 6.131 | Not recorded | 68fa77911e209131cea73f9f |
| SFT-SDXL | 5.8327 | Not recorded | 68fa77911e209131cea73f9f |
| DeDPO | 5.74 | Not recorded | 69896a26c97a0b3e3b5dc592 |
| DPO | 5.71 | Not recorded | 69896a26c97a0b3e3b5dc592 |
| DPO | 5.71 | Not recorded | 69896a26c97a0b3e3b5dc592 |
| DPO | 5.71 | Not recorded | 69896a26c97a0b3e3b5dc592 |
| DPO | 5.4 | Not recorded | 69896a26c97a0b3e3b5dc592 |
| DeDPO | 5.4 | Not recorded | 69896a26c97a0b3e3b5dc592 |
| DPO | 5.38 | Not recorded | 69896a26c97a0b3e3b5dc592 |
| DPO | 5.38 | Not recorded | 69896a26c97a0b3e3b5dc592 |
| SFT | 5.36 | Not recorded | 69896a26c97a0b3e3b5dc592 |
| SFT | 5.26 | Not recorded | 69896a26c97a0b3e3b5dc592 |

### InPO-SDXL

Aesthetic (Mean): 6.182. Result ID: 699540d93badfaffe62b85e2.

[Reporting paper](<https://www.sota2.com/research/paper/68fa77911e209131cea73f9f.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa77911e209131cea73f9f.md>)

[Implementation](<https://github.com/JaydenLyh/InPO>)

Recorded evaluation configuration: {"Base Model":"SDXL","Alignment Strategy":"InPO"}. This record does not assert independent reproduction.

### Base-SDXL

Aesthetic (Mean): 6.1346. Result ID: 699540d93badfaffe62b85df.

[Reporting paper](<https://www.sota2.com/research/paper/68fa77911e209131cea73f9f.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa70011e209131cea487fb.md>)

Recorded evaluation configuration: {"Base Model":"SDXL","Alignment Strategy":"None"}. This record does not assert independent reproduction.

### DPO-SDXL

Aesthetic (Mean): 6.131. Result ID: 699540d93badfaffe62b85e1.

[Reporting paper](<https://www.sota2.com/research/paper/68fa77911e209131cea73f9f.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa713d1e209131cea4ffe9.md>)

Recorded evaluation configuration: {"Base Model":"SDXL","Alignment Strategy":"DPO"}. This record does not assert independent reproduction.

### SFT-SDXL

Aesthetic (Mean): 5.8327. Result ID: 699540d93badfaffe62b85e0.

[Reporting paper](<https://www.sota2.com/research/paper/68fa77911e209131cea73f9f.md>)

Recorded evaluation configuration: {"Base Model":"SDXL","Alignment Strategy":"SFT"}. This record does not assert independent reproduction.

### DeDPO

Aesthetic (Mean): 5.74. Result ID: 699736985698665ec4866855.

[Reporting paper](<https://www.sota2.com/research/paper/69896a26c97a0b3e3b5dc592.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69896a26c97a0b3e3b5dc592.md>)

Recorded evaluation configuration: {"Model":"SDXL","# Pref. pairs":"1250","# Unpref. pairs":"3750","Configuration":"synthetic pref."}. This record does not assert independent reproduction.

### DPO

Aesthetic (Mean): 5.71. Result ID: 699736985698665ec4866852.

[Reporting paper](<https://www.sota2.com/research/paper/69896a26c97a0b3e3b5dc592.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa713d1e209131cea4ffe9.md>)

Recorded evaluation configuration: {"Model":"SDXL","# Pref. pairs":"1250","# Unpref. pairs":"0","Configuration":"25%"}. This record does not assert independent reproduction.

### DPO

Aesthetic (Mean): 5.71. Result ID: 699736985698665ec4866853.

[Reporting paper](<https://www.sota2.com/research/paper/69896a26c97a0b3e3b5dc592.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa713d1e209131cea4ffe9.md>)

Recorded evaluation configuration: {"Model":"SDXL","# Pref. pairs":"5000","# Unpref. pairs":"0","Configuration":"100%"}. This record does not assert independent reproduction.

### DPO

Aesthetic (Mean): 5.71. Result ID: 699736985698665ec4866854.

[Reporting paper](<https://www.sota2.com/research/paper/69896a26c97a0b3e3b5dc592.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa713d1e209131cea4ffe9.md>)

Recorded evaluation configuration: {"Model":"SDXL","# Pref. pairs":"1250","# Unpref. pairs":"3750","Configuration":"synthetic pref."}. This record does not assert independent reproduction.

### DPO

Aesthetic (Mean): 5.4. Result ID: 699736985698665ec486684f.

[Reporting paper](<https://www.sota2.com/research/paper/69896a26c97a0b3e3b5dc592.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa713d1e209131cea4ffe9.md>)

Recorded evaluation configuration: {"Model":"SD1.5","# Pref. pairs":"1250","# Unpref. pairs":"3750","Configuration":"synthetic pref."}. This record does not assert independent reproduction.

### DeDPO

Aesthetic (Mean): 5.4. Result ID: 699736985698665ec4866850.

[Reporting paper](<https://www.sota2.com/research/paper/69896a26c97a0b3e3b5dc592.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69896a26c97a0b3e3b5dc592.md>)

Recorded evaluation configuration: {"Model":"SD1.5","# Pref. pairs":"1250","# Unpref. pairs":"3750","Configuration":"synthetic pref."}. This record does not assert independent reproduction.

### DPO

Aesthetic (Mean): 5.38. Result ID: 699736985698665ec486684d.

[Reporting paper](<https://www.sota2.com/research/paper/69896a26c97a0b3e3b5dc592.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa713d1e209131cea4ffe9.md>)

Recorded evaluation configuration: {"Model":"SD1.5","# Pref. pairs":"1250","# Unpref. pairs":"0","Configuration":"25%"}. This record does not assert independent reproduction.

### DPO

Aesthetic (Mean): 5.38. Result ID: 699736985698665ec486684e.

[Reporting paper](<https://www.sota2.com/research/paper/69896a26c97a0b3e3b5dc592.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa713d1e209131cea4ffe9.md>)

Recorded evaluation configuration: {"Model":"SD1.5","# Pref. pairs":"5000","# Unpref. pairs":"0","Configuration":"100%"}. This record does not assert independent reproduction.

### SFT

Aesthetic (Mean): 5.36. Result ID: 699736985698665ec4866851.

[Reporting paper](<https://www.sota2.com/research/paper/69896a26c97a0b3e3b5dc592.md>)

Recorded evaluation configuration: {"Model":"SDXL","# Pref. pairs":"1250","# Unpref. pairs":"0"}. This record does not assert independent reproduction.

### SFT

Aesthetic (Mean): 5.26. Result ID: 699736985698665ec486684c.

[Reporting paper](<https://www.sota2.com/research/paper/69896a26c97a0b3e3b5dc592.md>)

Recorded evaluation configuration: {"Model":"SD1.5","# Pref. pairs":"1250","# Unpref. pairs":"0"}. This record does not assert independent reproduction.

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