Image-Text Retrieval on Flickr30K (I-T, T-I, Rsum)
87.4I->T Retrieval ScoreOurs‡
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Ours‡Params. (M)=11.8, Mem. (G)=25.42026.04 | 87.4 | 73.3 | 547 | |
| Ours†Params. (M)=12.9, Mem. (G)=25.52026.04 | 86.5 | 71.3 | 543.1 | |
| Ours♠Params. (M)=11.4, Mem. (G)=25.2, QPS=14.832026.04 | 86.4 | 71.6 | 543 | |
| SHERLParams. (M)=11.3, Mem. (G)=24.4, QPS=10.242026.04 | 86.1 | 71.1 | 542.3 | |
| SHERLParams. (M)=11.3, Mem. (G)=24.42026.04 | 86.1 | 71.1 | 542.3 | |
| Ours♡Params. (M)=12.4, Mem. (G)=25.1, QPS=14.712026.04 | 85.9 | 70.7 | 541.2 | |
| Fully-FTParams (M)=201.2, Training Memory (G)=22.1 * 8, Test Memory (G)=20.12, Architecture=VSE∞2023.08 | 85.6 | 73.3 | 546.6 | |
| Fully-FTBackbone=ResNeXt+BiGRU, Params. (M)=90.9, Mem. (G)=174.4, QPS=30.802026.04 | 85.6 | 70.2 | 539 | |
| Fully-FTParams. (M)=201.2, Mem. (G)=176.82026.04 | 85.6 | 73.3 | 546.6 | |
| Fully-FTParams. (M)=201.2, Mem. (G)=176.82026.04 | 85.6 | 73.3 | 546.6 | |
| MDPDBackbone=ResNeXt+BiGRU, Params. (M)=6.5, Mem. (G)=15.7, QPS=30.622026.04 | 85.3 | 69.8 | 539.1 | |
| UniPTParams (M)=12.4, Training Memory (G)=15.1, Test Memory (G)=20.19, Architecture=VSE∞2023.08 | 84.8 | 69.1 | 537.4 | |
| UniPTParams. (M)=12.4, Mem. (G)=24.4, QPS=8.612026.04 | 84.8 | 69.1 | 537.4 | |
| UniPTParams. (M)=12.4, Mem. (G)=24.42026.04 | 84.8 | 69.1 | 537.4 | |
| UniPTBackbone=ResNeXt+BiGRU, Params. (M)=6.4, Mem. (G)=15.0, QPS=17.832026.04 | 84 | 67.7 | 532.1 | |
| SSFBackbone=ResNeXt+BiGRU, Params. (M)=0.1, Mem. (G)=163.2, QPS=27.632026.04 | 83.7 | 66.8 | 528.5 | |
| BitFitBackbone=ResNeXt+BiGRU, Params. (M)=2.2, Mem. (G)=168.0, QPS=30.532026.04 | 83.4 | 67.4 | 530.6 | |
| LSTBackbone=ResNeXt+BiGRU, Params. (M)=2.3, Mem. (G)=15.0, QPS=13.862026.04 | 82.3 | 66.1 | 526.7 | |
| LSTParams (M)=9.7, Training Memory (G)=15.1, Test Memory (G)=20.31, Architecture=VSE∞2023.08 | 82.1 | 66.5 | 529.5 | |
| LSTParams. (M)=9.7, Mem. (G)=24.42026.04 | 82.1 | 66.5 | 529.5 | |
| LSTParams. (M)=9.7, Mem. (G)=24.42026.04 | 82.1 | 66.5 | 529.5 | |
| Partially⇓Params (M)=2.9, Training Memory (G)=15.0, Test Memory (G)=20.12, Architecture=VSE∞2023.08 | 75.6 | 59.8 | 508.3 | |
| PartiallyBackbone=ResNeXt+BiGRU, Params. (M)=2.1, Mem. (G)=14.9, QPS=31.262026.04 | 75.2 | 58.2 | 505.8 | |
| AdapterBackbone=ResNeXt+BiGRU, Params. (M)=3.5, Mem. (G)=176.8, QPS=28.652026.04 | 66.8 | 62.9 | 493.3 |