Image-to-Text Retrieval on Flickr30K (I→T@1)
90.8I->T@1HarmoCLIP
Evaluation Results
| Method | Links | |
|---|---|---|
| HarmoCLIPBackbone=ViT-B/16, Extra data / VLM augmentation=false2025.11 | 90.8 | |
| FG-CLIPBackbone=ViT-B/16, Extra data / VLM augmentation=true2025.11 | 90.7 | |
| EVA-CLIPBackbone=ViT-B/16, Extra data / VLM augmentation=false2025.11 | 85.7 | |
| Fully-FTBackbone=ResNeXt-101, Params. (M)=90.9, Memory (G)=21.8 * 82023.08 | 85.6 | |
| UniPTBackbone=ResNeXt-101, Params. (M)=6.4, Memory (G)=15.02023.08 | 84 | |
| SSFBackbone=ResNeXt-101, Params. (M)=0.1, Memory (G)=20.4 * 82023.08 | 83.7 | |
| BitFitBackbone=ResNeXt-101, Params. (M)=2.2, Memory (G)=21.0 * 82023.08 | 83.4 | |
| FineCLIPBackbone=ViT-B/16, Extra data / VLM augmentation=true2025.11 | 82.5 | |
| LSTBackbone=ResNeXt-101, Params. (M)=2.27, Memory (G)=15.02023.08 | 82.3 | |
| CLIPBackbone=ViT-B/16, Extra data / VLM augmentation=false2025.11 | 82.2 | |
| UniPTBackbone=BERT, Params. (M)=5.9, Memory (G)=3.12023.08 | 80.2 | |
| SSFBackbone=BERT, Params. (M)=0.2, Memory (G)=8.42023.08 | 80 | |
| AdaLoRABackbone=BERT, Params. (M)=1.0, Memory (G)=8.82023.08 | 79.8 | |
| Fully-FTBackbone=BERT, Params. (M)=109.5, Memory (G)=9.92023.08 | 79.7 | |
| FacTBackbone=BERT, Params. (M)=0.6, Memory (G)=8.72023.08 | 79.2 | |
| AdapterBackbone=BERT, Params. (M)=2.6, Memory (G)=8.82023.08 | 79.1 | |
| LoRABackbone=BERT, Params. (M)=1.1, Memory (G)=8.82023.08 | 78.8 | |
| PromptBackbone=BERT, Params. (M)=10.7, Memory (G)=9.42023.08 | 78.7 | |
| LSTBackbone=BERT, Params. (M)=7.5, Memory (G)=4.62023.08 | 77.9 | |
| BitFitBackbone=BERT, Params. (M)=0.9, Memory (G)=8.62023.08 | 77.3 | |
| PartiallyBackbone=ResNeXt-101, Params. (M)=2.1, Memory (G)=14.92023.08 | 75.2 | |
| PartiallyBackbone=BERT, Params. (M)=0.8, Memory (G)=1.02023.08 | 74.8 | |
| AdapterBackbone=ResNeXt-101, Params. (M)=3.5, Memory (G)=22.1 * 82023.08 | 66.8 | |
| CLIPSelfBackbone=ViT-B/16, Extra data / VLM augmentation=false2025.11 | 33.8 | |
| RegionCLIPBackbone=ViT-B/16, Extra data / VLM augmentation=false2025.11 | 3.9 |