Zero-Shot Sketch-Based Image Retrieval on Sketchy
0.723mAP@200CLIP for ZS-SBIR (Ours)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CLIP for ZS-SBIR (Ours)Backbone=ViT-B/16 (CLIP), Prompting Strategy=Fine-tuned Layer-Norm prompts2023.03 | 0.723 | 0.725 | |
| B-DeepBackbone=ViT-B/16 (CLIP), Prompting Strategy=Deep Prompting2023.03 | 0.702 | 0.718 | |
| B-IPBackbone=ViT-B/16 (CLIP), Prompting Strategy=Independent Prompting2023.03 | 0.691 | 0.711 | |
| B-MMBackbone=ViT-B/16 (CLIP), Prompting Strategy=Multi-modal Prompting2023.03 | 0.685 | 0.691 | |
| B-CondBackbone=ViT-B/16 (CLIP), Prompting Strategy=Conditional Prompting (ResNet-18)2023.03 | 0.618 | 0.675 | |
| ZS-Sketch3TEvaluation Protocol=Zero-Shot2023.03 | 0.579 | 0.648 | |
| ZS-GCNEvaluation Protocol=Zero-Shot2023.03 | 0.568 | 0.487 | |
| ZS-PSKD[VIT]Evaluation Protocol=Zero-Shot, Backbone=ViT2023.03 | 0.56 | 0.645 | |
| ZS-TVTEvaluation Protocol=Zero-Shot2023.03 | 0.531 | 0.618 | |
| ZS-TCNEvaluation Protocol=Zero-Shot2023.03 | 0.516 | 0.608 | |
| ZS-SAKEEvaluation Protocol=Zero-Shot2023.03 | 0.497 | 0.598 | |
| B-LinBackbone=ViT-B/16 (CLIP), Prompting Strategy=Linear Probe2023.03 | 0.422 | 0.512 | |
| ZS-IIAEEvaluation Protocol=Zero-Shot2023.03 | 0.373 | 0.485 | |
| ZS-GRLEvaluation Protocol=Zero-Shot2023.03 | 0.369 | 0.37 | |
| ZS-CVAEEvaluation Protocol=Zero-Shot2023.03 | 0.225 | 0.333 | |
| ZS-CAAEEvaluation Protocol=Zero-Shot2023.03 | 0.156 | 0.26 | |
| B-FTBackbone=ViT-B/16 (CLIP), Prompting Strategy=Fine-tune Image Encoder2023.03 | 0.102 | 0.166 |