Add dataset card, link to paper and code
#1
by nielsr HF Staff - opened
README.md
CHANGED
|
@@ -1,3 +1,17 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: mit
|
| 3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: mit
|
| 3 |
+
task_categories:
|
| 4 |
+
- other
|
| 5 |
+
tags:
|
| 6 |
+
- audio-visual
|
| 7 |
+
- video
|
| 8 |
+
- multimodal
|
| 9 |
+
---
|
| 10 |
+
|
| 11 |
+
# AV-SyncBench
|
| 12 |
+
|
| 13 |
+
[**Project Page**](https://fgt7t6g.github.io/AV-SyncBench) | [**Paper**](https://huggingface.co/papers/2607.00726) | [**Code**](https://github.com/fgt7t6g/AV-SyncBench)
|
| 14 |
+
|
| 15 |
+
AV-SyncBench is a benchmark designed to evaluate audio–visual synchronization representations by fully decoupling temporal consistency and semantic consistency.
|
| 16 |
+
|
| 17 |
+
Built from in-the-wild videos, it spans Voice, Music, and Sound across 10 scenarios and 5 challenge tasks. The benchmark contains 3,269 videos and 38,390 samples to evaluate and quantify feature quality for alignment and downstream tasks.
|