Image-to-Text Retrieval on COCO 1K (5-fold average test)
86.6R@1Ours†
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Ours†Backbone=ResNeXt-101 + BERT, CA (Cross-Attention)=false, Ensemble=true2022.11 | 86.6 | 98.2 | 99.4 | 549.9 | |
| OursBackbone=ResNeXt-101 + BERT, CA (Cross-Attention)=false, Ensemble=false2022.11 | 86.3 | 97.8 | 99.4 | 547.5 | |
| VSE∞†Backbone=ResNeXt-101 + BERT, CA (Cross-Attention)=false, Ensemble=true2022.11 | 85.6 | 98 | 99.4 | 548.1 | |
| VSE∞Backbone=ResNeXt-101 + BERT, CA (Cross-Attention)=false, Ensemble=false2022.11 | 84.5 | 98.1 | 99.4 | 545.4 | |
| Ours†Backbone=Faster R-CNN + Bi-GRU, CA (Cross-Attention)=false, Ensemble=true2022.11 | 80.6 | 96.3 | 98.8 | 528 | |
| NAAFBackbone=Faster R-CNN + Bi-GRU, CA (Cross-Attention)=true, Ensemble=true2022.11 | 80.5 | 96.5 | 98.8 | 527.2 | |
| OursBackbone=Faster R-CNN + Bi-GRU, CA (Cross-Attention)=false, Ensemble=false2022.11 | 79.8 | 96.2 | 98.6 | 524.6 | |
| SGRAFBackbone=Faster R-CNN + Bi-GRU, CA (Cross-Attention)=true, Ensemble=true2022.11 | 79.6 | 96.2 | 98.5 | 524.3 | |
| VSE∞Backbone=Faster R-CNN + Bi-GRU, CA (Cross-Attention)=false, Ensemble=false2022.11 | 78.5 | 96 | 98.7 | 520.8 | |
| IMRAMBackbone=Faster R-CNN + Bi-GRU, CA (Cross-Attention)=true, Ensemble=true2022.11 | 76.7 | 95.6 | 98.5 | 516.6 | |
| VSRNBackbone=Faster R-CNN + Bi-GRU, CA (Cross-Attention)=false, Ensemble=true2022.11 | 76.2 | 94.8 | 98.2 | 516.8 | |
| CAANBackbone=Faster R-CNN + Bi-GRU, CA (Cross-Attention)=true, Ensemble=false2022.11 | 75.5 | 95.4 | 98.5 | 515.6 | |
| SCANBackbone=Faster R-CNN + Bi-GRU, CA (Cross-Attention)=true, Ensemble=true2022.11 | 72.7 | 94.8 | 98.4 | 507.9 | |
| OursBackbone=ResNet-152 + Bi-GRU, CA (Cross-Attention)=false, Ensemble=false2022.11 | 70.3 | 91.5 | 96.3 | 493.2 | |
| PVSEBackbone=ResNet-152 + Bi-GRU, CA (Cross-Attention)=false, Ensemble=false2022.11 | 69.2 | 91.6 | 96.6 | 492.8 | |
| PCMEBackbone=ResNet-152 + Bi-GRU, CA (Cross-Attention)=false, Ensemble=false2022.11 | 68.8 | — | — | — | |
| VSE++Backbone=ResNet-152 + Bi-GRU, CA (Cross-Attention)=false, Ensemble=false2022.11 | 64.6 | 90 | 95.7 | 478.6 |