Datasets:
Add overview picture
Browse files
README.md
CHANGED
|
@@ -17,6 +17,15 @@ size_categories:
|
|
| 17 |
|
| 18 |
ArenaBench is a large-scale benchmark designed for comprehensive evaluation of deepfake detection methods. It contains approximately 45K testing samples and evaluates detectors across five levels, including in-domain, cross-model, cross-manipulation, commercial models, and real-world scenarios.
|
| 19 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 20 |
## Benchmark Design
|
| 21 |
|
| 22 |
ArenaBench evaluates deepfake detectors under five complementary settings:
|
|
|
|
| 17 |
|
| 18 |
ArenaBench is a large-scale benchmark designed for comprehensive evaluation of deepfake detection methods. It contains approximately 45K testing samples and evaluates detectors across five levels, including in-domain, cross-model, cross-manipulation, commercial models, and real-world scenarios.
|
| 19 |
|
| 20 |
+
## Overview
|
| 21 |
+
<p align="center">
|
| 22 |
+
<img src="assets/arenabench_overview.png" width="900">
|
| 23 |
+
</p>
|
| 24 |
+
|
| 25 |
+
<p align="center">
|
| 26 |
+
<em>Overview of the ArenaBench benchmark.</em>
|
| 27 |
+
</p>
|
| 28 |
+
|
| 29 |
## Benchmark Design
|
| 30 |
|
| 31 |
ArenaBench evaluates deepfake detectors under five complementary settings:
|