Datasets:
Update dataset card
Browse files
README.md
CHANGED
|
@@ -236,21 +236,23 @@ export GEMINI_API_KEY=<your-gemini-key>
|
|
| 236 |
python evaluation/agentedit_bench_score.py \
|
| 237 |
--bench_dir "$BENCH" \
|
| 238 |
--results_dir "$PWD/results/aurora" \
|
| 239 |
-
--api
|
|
|
|
|
|
|
| 240 |
--model gemini-2.5-pro
|
| 241 |
```
|
| 242 |
|
| 243 |
-
|
|
|
|
|
|
|
| 244 |
`summary.by_edit_type.<edit_type>.avg_pct_on_1` and `summary.overall_pct_on_1`,
|
| 245 |
-
multiplied by 100, are the numbers reported in the table below.
|
| 246 |
-
OpenAI-compatible judge can be used instead with
|
| 247 |
-
`--api openai --base_url <url> --model <model>`.
|
| 248 |
|
| 249 |
## Evaluation protocol
|
| 250 |
|
| 251 |
-
Gemini 2.5 Pro at temperature 0
|
| 252 |
-
case, taken at frames 0, *k*, and 2*k* of the
|
| 253 |
-
*k* = ⌊N/3⌋, and scores each axis from 0 to 3.
|
| 254 |
|
| 255 |
| # | Axis | What it measures | Applies to |
|
| 256 |
|---|---|---|---|
|
|
@@ -303,10 +305,10 @@ were created and curated by the Aurora authors.
|
|
| 303 |
## Citation
|
| 304 |
|
| 305 |
```bibtex
|
| 306 |
-
@
|
| 307 |
title={Aurora: Unified Video Editing with a Tool-Using Agent},
|
| 308 |
author={Yu, Yongsheng and Zeng, Ziyun and Xiao, Zhiyuan and Zhou, Zhenghong and Hua, Hang and Xiong, Wei and Luo, Jiebo},
|
| 309 |
-
|
| 310 |
year={2026}
|
| 311 |
}
|
| 312 |
```
|
|
|
|
| 236 |
python evaluation/agentedit_bench_score.py \
|
| 237 |
--bench_dir "$BENCH" \
|
| 238 |
--results_dir "$PWD/results/aurora" \
|
| 239 |
+
--api openai \
|
| 240 |
+
--base_url https://generativelanguage.googleapis.com/v1beta/openai/ \
|
| 241 |
+
--api_key "$GEMINI_API_KEY" \
|
| 242 |
--model gemini-2.5-pro
|
| 243 |
```
|
| 244 |
|
| 245 |
+
This runs the judge as in the paper: Gemini 2.5 Pro at temperature 0, called
|
| 246 |
+
through its OpenAI-compatible endpoint, without web search. Scores are written
|
| 247 |
+
to `<results_dir>/scores.json`. The values
|
| 248 |
`summary.by_edit_type.<edit_type>.avg_pct_on_1` and `summary.overall_pct_on_1`,
|
| 249 |
+
multiplied by 100, are the numbers reported in the table below.
|
|
|
|
|
|
|
| 250 |
|
| 251 |
## Evaluation protocol
|
| 252 |
|
| 253 |
+
Gemini 2.5 Pro at temperature 0, without web search, compares three
|
| 254 |
+
BEFORE/AFTER frame pairs per case, taken at frames 0, *k*, and 2*k* of the
|
| 255 |
+
edited clip with *k* = ⌊N/3⌋, and scores each axis from 0 to 3.
|
| 256 |
|
| 257 |
| # | Axis | What it measures | Applies to |
|
| 258 |
|---|---|---|---|
|
|
|
|
| 305 |
## Citation
|
| 306 |
|
| 307 |
```bibtex
|
| 308 |
+
@inproceedings{yu2026aurora,
|
| 309 |
title={Aurora: Unified Video Editing with a Tool-Using Agent},
|
| 310 |
author={Yu, Yongsheng and Zeng, Ziyun and Xiao, Zhiyuan and Zhou, Zhenghong and Hua, Hang and Xiong, Wei and Luo, Jiebo},
|
| 311 |
+
booktitle={Advances in Neural Information Processing Systems (NeurIPS)},
|
| 312 |
year={2026}
|
| 313 |
}
|
| 314 |
```
|