Text Retrieval on COCO (clean/ASR)
72.9Clean AccuracyCLIP
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CLIPBackbone=ViT-B/32, Attack Type=BadCLIP2024.12 | 72.9 | 96.8 | |
| CLIPBackbone=ViT-B/32, Attack Type=Blended, Trigger Pattern=Stripes2024.12 | 72.6 | 99.8 | |
| CLIPBackbone=ViT-B/32, Attack Type=BadNet, Trigger Pattern=Stripes2024.12 | 72.5 | 98.9 | |
| CLIPBackbone=ViT-B/32, Attack Type=BadNet, Trigger Pattern=Rand.2024.12 | 72.2 | 85.6 | |
| CLIPBackbone=ViT-B/32, Attack Type=Blended, Trigger Pattern=Triangles2024.12 | 71.9 | 99.9 | |
| CLIPBackbone=ViT-B/32, Attack Type=Blended, Trigger Pattern=Text2024.12 | 71.8 | 99.8 | |
| CLIPBackbone=ViT-B/32, Attack Type=Blended, Trigger Pattern=Rand.2024.12 | 71.3 | 59.9 | |
| CleanCLIPBackbone=ViT-B/32, Attack Type=Blended, Trigger Pattern=Stripes2024.12 | 70.6 | 22.4 | |
| CleanCLIPBackbone=ViT-B/32, Attack Type=Blended, Trigger Pattern=Text2024.12 | 70.6 | 55.4 | |
| PARBackbone=ViT-B/32, Attack Type=BadCLIP2024.12 | 70.6 | 27.1 | |
| CleanCLIPBackbone=ViT-B/32, Attack Type=BadNet, Trigger Pattern=Stripes2024.12 | 70.5 | 84.7 | |
| CleanCLIPBackbone=ViT-B/32, Attack Type=Blended, Trigger Pattern=Rand.2024.12 | 70.4 | 2.9 | |
| CleanCLIPBackbone=ViT-B/32, Attack Type=BadNet, Trigger Pattern=Rand.2024.12 | 70.3 | 11.2 | |
| CleanCLIPBackbone=ViT-B/32, Attack Type=Blended, Trigger Pattern=Triangles2024.12 | 70.2 | 98.5 | |
| PARBackbone=ViT-B/32, Attack Type=Blended, Trigger Pattern=Text2024.12 | 70.2 | 45.4 | |
| CleanCLIPBackbone=ViT-B/32, Attack Type=BadCLIP2024.12 | 70.1 | 28.5 | |
| PARBackbone=ViT-B/32, Attack Type=Blended, Trigger Pattern=Triangles2024.12 | 70 | 91.5 | |
| PARBackbone=ViT-B/32, Attack Type=BadNet, Trigger Pattern=Stripes2024.12 | 69.5 | 51.6 | |
| PARBackbone=ViT-B/32, Attack Type=BadNet, Trigger Pattern=Rand.2024.12 | 69.4 | 1.4 | |
| PARBackbone=ViT-B/32, Attack Type=Blended, Trigger Pattern=Rand.2024.12 | 69.4 | 1.5 | |
| PARBackbone=ViT-B/32, Attack Type=Blended, Trigger Pattern=Stripes2024.12 | 69.4 | 1.7 |