# Human attention prediction on BDD-A DKL \> 2 non-trivial frames (test)

Source: [Human attention prediction on BDD-A DKL \> 2 non-trivial frames (test)](<https://www.sota2.com/research/sota/human-attention-prediction-on-bdd-a-dkl-2-non-trivial-frames-test>)

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

Data updated: 2026-02-26T05:52:08.805Z

Evaluation of human attention prediction models using the specific subset of the BDD-A dataset defined by DKL \> 2 non-trivial frames (test split), typically measured using KL divergence and correlation coefficients.

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 | Human attention prediction |
| Dataset (including any stated split) | BDD-A DKL \> 2 non-trivial frames (test) |
| Metric | Mean KL Divergence |
| Unit | SCALAR |
| Direction | Lower is better |
| Results loaded | 5 of 5 |
| 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

- [Mean KL Divergence](<https://www.sota2.com/research/sota/human-attention-prediction-on-bdd-a-dkl-2-non-trivial-frames-test.md?metric=Mean%20KL%20Divergence>): Lower is better; unit: SCALAR.

- [KL Divergence (95% CI)](<https://www.sota2.com/research/sota/human-attention-prediction-on-bdd-a-dkl-2-non-trivial-frames-test.md?metric=KL%20Divergence%20(95%25%20CI)>): Lower is better; unit: SCALAR.

- [Mean Correlation Coefficient](<https://www.sota2.com/research/sota/human-attention-prediction-on-bdd-a-dkl-2-non-trivial-frames-test.md?metric=Mean%20Correlation%20Coefficient>): Higher is better; unit: SCALAR.

- [Correlation Coefficient (95% CI)](<https://www.sota2.com/research/sota/human-attention-prediction-on-bdd-a-dkl-2-non-trivial-frames-test.md?metric=Correlation%20Coefficient%20(95%25%20CI)>): Higher is better; unit: SCALAR.

## Recorded results

| Method | Mean KL Divergence | Result date | Report paper |
| --- | --- | --- | --- |
| BDD-A prediction model (HWS) | 1.67 | Not recorded | 68fa651c1e209131cea08683 |
| BDD-A prediction model (default) | 1.71 | Not recorded | 68fa651c1e209131cea08683 |
| SALICON | 1.76 | Not recorded | 68fa651c1e209131cea08683 |
| Averaged attention baseline | 1.87 | Not recorded | 68fa651c1e209131cea08683 |
| DR(eye)VE | 2.63 | Not recorded | 68fa651c1e209131cea08683 |

### BDD-A prediction model (HWS)

Mean KL Divergence: 1.67. Result ID: 698c9dc00b614d12cff1f80e.

[Reporting paper](<https://www.sota2.com/research/paper/68fa651c1e209131cea08683.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa651c1e209131cea08683.md>)

Recorded evaluation configuration: {"Sampling Strategy":"Human Weighted Sampling (HWS)","Output Size":"64x36"}. This record does not assert independent reproduction.

### BDD-A prediction model (default)

Mean KL Divergence: 1.71. Result ID: 698c9dc00b614d12cff1f80d.

[Reporting paper](<https://www.sota2.com/research/paper/68fa651c1e209131cea08683.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa651c1e209131cea08683.md>)

Recorded evaluation configuration: {"Sampling Strategy":"default","Output Size":"64x36"}. This record does not assert independent reproduction.

### SALICON

Mean KL Divergence: 1.76. Result ID: 698c9dc00b614d12cff1f80b.

[Reporting paper](<https://www.sota2.com/research/paper/68fa651c1e209131cea08683.md>)

Recorded evaluation configuration: {"Fine-tuned":"true","Implementation":"Open source \[26\]","Output Size":"64x36 (scaled)"}. This record does not assert independent reproduction.

### Averaged attention baseline

Mean KL Divergence: 1.87. Result ID: 698c9dc00b614d12cff1f80a.

[Reporting paper](<https://www.sota2.com/research/paper/68fa651c1e209131cea08683.md>)

Recorded evaluation configuration: {"Description":"Predicts the averaged human attention map of training videos"}. This record does not assert independent reproduction.

### DR(eye)VE

Mean KL Divergence: 2.63. Result ID: 698c9dc00b614d12cff1f80c.

[Reporting paper](<https://www.sota2.com/research/paper/68fa651c1e209131cea08683.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa63531e209131ce9fe2b9.md>)

Recorded evaluation configuration: {"Pre-trained":"true","Fine-tuned":"false","Output Size":"64x36 (scaled)"}. This record does not assert independent reproduction.

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