# Semantic Attribute Prediction and Counterfactual Recourse on CELEBA-ATTRIBUTES

Source: [Semantic Attribute Prediction and Counterfactual Recourse on CELEBA-ATTRIBUTES](<https://www.sota2.com/research/sota/semantic-attribute-prediction-and-counterfactual-recourse-on-celeba-attributes>)

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

Data updated: 2026-05-19T09:33:23.231Z

Evaluation of models performing semantic attribute prediction and counterfactual recourse analysis on the CELEBA-ATTRIBUTES dataset, reporting accuracy (Acc.), Variance Difference (VD), Counterfactual Recourse Gap (CRG), and Directional Similarity (DirSim).

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 | Semantic Attribute Prediction and Counterfactual Recourse |
| Dataset (including any stated split) | CELEBA-ATTRIBUTES |
| Metric | Accuracy |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 7 of 7 |
| 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

- [Accuracy](<https://www.sota2.com/research/sota/semantic-attribute-prediction-and-counterfactual-recourse-on-celeba-attributes.md?metric=Accuracy>): Higher is better; unit: PERCENT.

- [Variance Difference (VD)](<https://www.sota2.com/research/sota/semantic-attribute-prediction-and-counterfactual-recourse-on-celeba-attributes.md?metric=Variance%20Difference%20(VD)>): Lower is better; unit: SCALAR.

- [Counterfactual Recourse Gap (CRG)](<https://www.sota2.com/research/sota/semantic-attribute-prediction-and-counterfactual-recourse-on-celeba-attributes.md?metric=Counterfactual%20Recourse%20Gap%20(CRG)>): Lower is better; unit: SCALAR.

- [Directional Similarity (DirSim)](<https://www.sota2.com/research/sota/semantic-attribute-prediction-and-counterfactual-recourse-on-celeba-attributes.md?metric=Directional%20Similarity%20(DirSim)>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Accuracy | Result date | Report paper |
| --- | --- | --- | --- |
| PTQ INT8 | 90.2 | Not recorded | 6a0bee90dbc836b87d82c3a9 |
| CFQ-QAT | 89.9 | Not recorded | 6a0bee90dbc836b87d82c3a9 |
| CF-PTQ | 89.8 | Not recorded | 6a0bee90dbc836b87d82c3a9 |
| MixedPrec-QAT | 89.7 | Not recorded | 6a0bee90dbc836b87d82c3a9 |
| KD+Quant | 89.5 | Not recorded | 6a0bee90dbc836b87d82c3a9 |
| PTQ INT4 | 88.6 | Not recorded | 6a0bee90dbc836b87d82c3a9 |
| Prune+Quant | 88.3 | Not recorded | 6a0bee90dbc836b87d82c3a9 |

### PTQ INT8

Accuracy: 90.2. Result ID: 6a0c2e54fbc373592c029ceb.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee90dbc836b87d82c3a9.md>)

Recorded evaluation configuration: {"Overhead":"1.00×"}. This record does not assert independent reproduction.

### CFQ-QAT

Accuracy: 89.9. Result ID: 6a0c2e54fbc373592c029cf1.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee90dbc836b87d82c3a9.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a0bee90dbc836b87d82c3a9.md>)

Recorded evaluation configuration: {"Overhead":"1.46×"}. This record does not assert independent reproduction.

### CF-PTQ

Accuracy: 89.8. Result ID: 6a0c2e54fbc373592c029cf0.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee90dbc836b87d82c3a9.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a0bee90dbc836b87d82c3a9.md>)

Recorded evaluation configuration: {"Overhead":"1.28×"}. This record does not assert independent reproduction.

### MixedPrec-QAT

Accuracy: 89.7. Result ID: 6a0c2e54fbc373592c029ced.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee90dbc836b87d82c3a9.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa65a41e209131cea0b895.md>)

Recorded evaluation configuration: {"Overhead":"1.18×"}. This record does not assert independent reproduction.

### KD+Quant

Accuracy: 89.5. Result ID: 6a0c2e54fbc373592c029cef.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee90dbc836b87d82c3a9.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa61b81e209131ce9f5b87.md>)

Recorded evaluation configuration: {"Overhead":"1.25×"}. This record does not assert independent reproduction.

### PTQ INT4

Accuracy: 88.6. Result ID: 6a0c2e54fbc373592c029cec.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee90dbc836b87d82c3a9.md>)

Recorded evaluation configuration: {"Overhead":"1.00×"}. This record does not assert independent reproduction.

### Prune+Quant

Accuracy: 88.3. Result ID: 6a0c2e54fbc373592c029cee.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee90dbc836b87d82c3a9.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa624a1e209131ce9f8a38.md>)

Recorded evaluation configuration: {"Overhead":"1.13×"}. This record does not assert independent reproduction.

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