Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Safety Alignment on Visual Adversarial Attacks

43.1ASR

Vanilla

-0.68410.68322.0533.417Oct 15, 2025
Updated 19d ago

Evaluation Results

MethodLinks
2025.10
43.1
2025.10
37.1
2025.10
34.5
2025.10
26.4
2025.10
23.4
2025.10
20.3
2025.10
20.3
2025.10
19.8
2025.10
18.6
2025.10
17.4
2025.10
16.8
2025.10
16.3
2025.10
16.1
2025.10
15.9
2025.10
14.3
2025.10
13.7
2025.10
13.5
2025.10
12.1
2025.10
11
2025.10
10.6
2025.10
10
2025.10
9.5
2025.10
8.4
2025.10
7
2025.10
6.9
2025.10
6.8
2025.10
6.8
2025.10
6.1
2025.10
5.6
2025.10
5.4
2025.10
5.4
2025.10
5.3
2025.10
4.8
2025.10
4.5
2025.10
4.2
2025.10
3.6
2025.10
2.9
2025.10
1.8
2025.10
1.4
2025.10
1