# Video Question Answering on NExT-QA Hard Split - Causal

Source: [Video Question Answering on NExT-QA Hard Split - Causal](<https://www.sota2.com/research/sota/video-question-answering-on-next-qa-hard-split-causal>)

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

Data updated: 2026-02-26T05:28:46.468Z

Evaluation of model performance on the Causal subset of the NExT-QA Hard Split for Video Question Answering, reporting 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 | Video Question Answering |
| Dataset (including any stated split) | NExT-QA Hard Split - Causal |
| Metric | Accuracy |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 3 of 3 |
| 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/video-question-answering-on-next-qa-hard-split-causal.md?metric=Accuracy>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Accuracy | Result date | Report paper |
| --- | --- | --- | --- |
| ViperGPT | 56.4 | Not recorded | 68fa6ec31e209131cea41d4d |
| HiTeA | 47.8 | Not recorded | 68fa6ec31e209131cea41d4d |
| ATP | 43.3 | Not recorded | 68fa6ec31e209131cea41d4d |

### ViperGPT

Accuracy: 56.4. Result ID: 699056443a6c314849ffd119.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6ec31e209131cea41d4d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6ec31e209131cea41d4d.md>)

[Implementation](<https://www.sota2.com/viper.cs.columbia.edu/>)

Recorded evaluation configuration: {"Supervision":"No"}. This record does not assert independent reproduction.

### HiTeA

Accuracy: 47.8. Result ID: 699056443a6c314849ffd118.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6ec31e209131cea41d4d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6e4b1e209131cea3f143.md>)

Recorded evaluation configuration: {"Supervision":"Yes"}. This record does not assert independent reproduction.

### ATP

Accuracy: 43.3. Result ID: 699056443a6c314849ffd117.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6ec31e209131cea41d4d.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6cba1e209131cea359da.md>)

[Implementation](<https://stanfordvl.github.io/atp-revisit-video-lang/>)

Recorded evaluation configuration: {"Supervision":"Yes"}. This record does not assert independent reproduction.

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