Image-to-Text Retrieval on RSITMD (test)
32.3R@1GeoRSCLIP-FT
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| GeoRSCLIP-FTParadigm=GVLM + FT, Tuned on=RS5M + RET-22023.06 | 32.3 | 53.32 | 67.92 | — | — | |
| GeoRSCLIP-FTParadigm=GVLM + FT, Tuned on=RS5M + RSITMD2023.06 | 30.09 | 51.55 | 63.27 | — | — | |
| RemoteCLIPParadigm=GVLM + FT, Tuned on=RET-3 + DET-10 + SEG-42023.06 | 27.88 | 50.66 | 65.71 | — | — | |
| RemoteCLIPBackbone=ViT-B/322026.07 | 27.88 | 50.66 | 65.71 | — | 296.29 | |
| ELCBackbone=ViT-B/322026.07 | 27.43 | 51.11 | 63.72 | — | 294.07 | |
| ELC0Backbone=ViT-B/322026.07 | 27.21 | 50.22 | 63.27 | — | 291.84 | |
| GLGSABackbone=ViT-B/322026.07 | 26.77 | 48.23 | 60.39 | — | 279.76 | |
| FGVLABackbone=ViT-B/322026.07 | 26.33 | 51.99 | 64.38 | — | 291.51 | |
| Full fine-tuningReference=-, Backbone (image/text)=CLIP(ViT-B-32), Params=151M2023.08 | 24.16 | 47.12 | 61.28 | 45.33 | — | |
| AIRBackbone=ViT-B/322026.07 | 24.12 | 42.04 | 54.2 | — | 258.67 | |
| OursReference=-, Backbone (image/text)=CLIP(ViT-B-32), Params=0.16M2023.08 | 23.67 | 44.07 | 60.36 | 44.47 | — | |
| PE-RSITRParadigm=GVLM + FT, Tuned on=RSITMD2023.06 | 23.67 | 44.07 | 60.36 | — | — | |
| PE-RSITRBackbone=ViT-B/322026.07 | 23.67 | 44.07 | 60.36 | — | 266.8 | |
| PAUBackbone=ViT-B/322026.07 | 22.12 | 41.37 | 54.87 | — | 250.39 | |
| CUPBackbone=ViT-B/322026.07 | 21.57 | 43.7 | 56.75 | — | 252.94 | |
| GeoAlignCLIPBackbone=ViT-L/142026.03 | 21.02 | 37.17 | 49.34 | — | — | |
| PolyBackbone=ViT-B/322026.07 | 21.02 | 35.84 | 47.79 | — | 220.62 | |
| VSE∞Backbone=ViT-B/322026.07 | 21.02 | 46.02 | 57.74 | — | 250.09 | |
| LRSCLIPBackbone=ViT-B/162026.03 | 20.58 | 40.71 | 50.66 | — | — | |
| LRSCLIPBackbone=ViT-L/142026.03 | 20.35 | 39.82 | 51.77 | — | — | |
| GeoRSCLIPBackbone=ViT-L/14-3362026.03 | 20.13 | 37.17 | 48.23 | — | — | |
| UniAdapterReference=arXiv’23, Backbone (image/text)=CLIP(ViT-B-32), Params=0.55M2023.08 | 19.86 | 36.32 | 51.28 | 39.23 | — | |
| Single LanguageReference=JSTARS’22, Backbone (image/text)=CLIP(ViT-B-32), Params=151M2023.08 | 19.69 | 40.26 | 54.42 | 41.38 | — | |
| MultilanguageParadigm=Supervised, Tuned on=RSITMD2023.06 | 19.69 | 40.26 | 54.42 | — | — | |
| GeoRSCLIPBackbone=ViT-L/142026.03 | 19.69 | 38.5 | 48.89 | — | — | |
| VSE++Backbone=ViT-B/322026.07 | 19.69 | 38.5 | 52.21 | — | 223.54 | |
| InfoNCEBackbone=ViT-B/322026.07 | 19.47 | 39.6 | 53.32 | — | 245.14 | |
| GeoRSCLIPParadigm=GVLM + FT, Tuned on=RS5M2023.06 | 19.03 | 34.51 | 46.46 | — | — | |
| SIRSBackbone=ViT-B/322026.07 | 18.45 | 44.06 | 57.73 | — | 241.26 | |
| GeoAlignCLIPBackbone=ViT-B/162026.03 | 18.36 | 34.29 | 48.67 | — | — | |
| Cross-Modal AdapterReference=arXiv’22, Backbone (image/text)=CLIP(ViT-B-32), Params=0.16M2023.08 | 18.16 | 36.08 | 48.72 | 38.06 | — | |
| PIRParadigm=Supervised, Tuned on=RSITMD2023.06 | 18.14 | 41.15 | 52.88 | — | — | |
| MTGFEParadigm=Supervised, Tuned on=RSITMD2023.06 | 17.92 | 40.93 | 53.32 | — | — | |
| SMLGN2026.07 | 17.26 | 39.38 | 51.55 | — | 225.72 | |
| Linear probeReference=ICML’21, Backbone (image/text)=CLIP(ViT-B-32), Params=1.05M2023.08 | 17.02 | 33.12 | 48.35 | 36.25 | — | |
| GeoRSCLIPParadigm=GVLM + Pfeiffer, Tuned on=RS5M2023.06 | 16.81 | 34.73 | 44.69 | — | — | |
| DOVEParms=38 M2023.10 | 16.81 | 36.8 | 49.93 | — | — | |
| KAMCL2026.07 | 16.81 | 34.96 | 47.12 | — | 215.21 | |
| AdaptFormerReference=NIPS’22, Backbone (image/text)=CLIP(ViT-B-32), Params=0.17M2023.08 | 16.71 | 30.16 | 42.91 | 34.81 | — | |
| GeoRSCLIPParadigm=GVLM + LoRA, Tuned on=RS5M2023.06 | 16.37 | 32.96 | 42.7 | — | — | |
| FAAMIParadigm=Supervised, Tuned on=RSITMD2023.06 | 16.15 | 35.62 | 48.89 | — | — | |
| GeoRSCLIPParadigm=GVLM + UniPELT, Tuned on=RS5M2023.06 | 16.15 | 31.86 | 42.04 | — | — | |
| FGCLIPBackbone=ViT-L/14-3362026.03 | 16.15 | 33.41 | 45.13 | — | — | |
| ConvpassReference=arXiv’22, Backbone (image/text)=CLIP(ViT-B-32), Params=0.14M2023.08 | 16.03 | 30.16 | 40.26 | 32.55 | — | |
| GeoRSCLIPParadigm=GVLM + Prefixtuning, Tuned on=RS5M2023.06 | 15.93 | 31.86 | 41.81 | — | — | |
| DOVE-SParms=25 M2023.10 | 15.71 | 31.86 | 44.69 | — | — | |
| GeoRSCLIPBackbone=ViT-B/322026.03 | 15.49 | 32.96 | 43.58 | — | — | |
| VPTReference=ECCV’22, Backbone (image/text)=CLIP(ViT-B-32), Params=0.46M2023.08 | 14.98 | 32.05 | 40.15 | 34.14 | — | |
| GaLR with MRReference=TGRS’22, Backbone (image/text)=ResNet18, biGRU, Params=46.89M2023.08 | 14.82 | 31.64 | 42.48 | 31.41 | — | |
| GaLRParadigm=Supervised, Tuned on=RSITMD2023.06 | 14.82 | 31.64 | 42.48 | — | — | |
| GaLRParms=46 M2023.10 | 14.82 | 31.64 | 42.48 | — | — | |
| GaLR2026.07 | 14.82 | 31.64 | 42.48 | — | 188.45 | |
| AdapterReference=ICML’19, Backbone (image/text)=CLIP(ViT-B-32), Params=0.17M2023.08 | 13.75 | 27.64 | 39.96 | 32.37 | — | |
| FGCLIPBackbone=ViT-B/162026.03 | 13.72 | 28.76 | 36.95 | — | — | |
| SWANParadigm=Supervised, Tuned on=RSITMD2023.06 | 13.35 | 32.15 | 46.9 | — | — | |
| SWANParms=40 M2023.10 | 13.35 | 32.15 | 46.9 | — | — | |
| MCRNReference=JAG’22, Backbone (image/text)=ResNet18, biGRU, Params=52.35M2023.08 | 13.27 | 29.42 | 41.59 | 30.33 | — | |
| HVSAParadigm=Supervised, Tuned on=RSITMD2023.06 | 13.2 | 32.08 | 45.58 | — | — | |
| HVSA2023.10 | 13.2 | 32.08 | 45.58 | — | — | |
| GaLR w/o MRReference=TGRS’22, Backbone (image/text)=ResNet18, biGRU, Params=46.89M2023.08 | 13.05 | 30.09 | 42.7 | 31 | — | |
| CLIP-AdapterReference=arXiv’21, Backbone (image/text)=CLIP(ViT-B-32), Params=0.52M2023.08 | 12.83 | 28.84 | 39.05 | 32.38 | — | |
| SkyCLIPBackbone=ViT-L/14-pct302026.03 | 12.83 | 28.32 | 39.82 | — | — | |
| VL-PromptReference=-, Backbone (image/text)=CLIP(ViT-B-32), Params=0.47M2023.08 | 12.81 | 31.28 | 42.64 | 32.4 | — | |
| CoOpReference=IJCV’22, Backbone (image/text)=CLIP(ViT-B-32), Params=0.10M2023.08 | 12.19 | 30.69 | 42.82 | 30.51 | — | |
| SkyCLIPBackbone=ViT-L/14-pct502026.03 | 12.17 | 28.32 | 38.5 | — | — | |
| CAMP-tripletReference=ICCV’19, Backbone (image/text)=Faster R-CNN, biGRU, Params=27.03M2023.08 | 11.73 | 26.99 | 38.05 | 26.2 | — | |
| SCAN i-tReference=ECCV’18, Backbone (image/text)=Faster R-CNN, biGRU, Params=13.68M2023.08 | 11.06 | 25.88 | 39.38 | 26.28 | — | |
| AMFMN-softReference=TGRS’22, Backbone (image/text)=ResNet18, biGRU, Params=35.94M2023.08 | 11.06 | 25.88 | 39.82 | 28.74 | — | |
| AMFMN-fusionReference=TGRS’22, Backbone (image/text)=ResNet18, biGRU, Params=35.94M2023.08 | 11.06 | 29.2 | 38.72 | 29.32 | — | |
| AFMFNParadigm=Supervised, Tuned on=RSITMD2023.06 | 11.06 | 29.2 | 38.72 | — | — | |
| LongCLIPBackbone=ViT-L/142026.03 | 11.06 | 27.43 | 38.27 | — | — | |
| AMFMN-simReference=TGRS’22, Backbone (image/text)=ResNet18, biGRU, Params=35.94M2023.08 | 10.63 | 24.78 | 41.81 | 29.72 | — | |
| AMFMNParms=36 M2023.10 | 10.63 | 24.78 | 41.81 | — | — | |
| AMFMN2026.07 | 10.63 | 24.78 | 41.81 | — | 178.29 | |
| CLIPBackbone=ViT-L/142026.03 | 10.62 | 27.43 | 36.73 | — | — | |
| MTFNReference=ACM MM’19, Backbone (image/text)=Faster R-CNN, biGRU, Params=77.90M2023.08 | 10.4 | 27.65 | 36.28 | 26.92 | — | |
| VSE++Reference=BMVC’18, Backbone (image/text)=ResNet18, GRU, Params=15.78M2023.08 | 10.38 | 27.65 | 39.6 | 24.83 | — | |
| VSE++Paradigm=Supervised, Tuned on=RSITMD2023.06 | 10.38 | 27.65 | 39.6 | — | — | |
| SCAN t-iReference=ECCV’18, Backbone (image/text)=Faster R-CNN, biGRU, Params=13.68M2023.08 | 10.18 | 28.53 | 38.49 | 26.64 | — | |
| LW-MCR-dReference=TGRS’22, Backbone (image/text)=SqueezeNet, -, Params=1.65M2023.08 | 10.18 | 28.98 | 39.82 | 27.79 | — | |
| LW-MCRParadigm=Supervised, Tuned on=RSITMD2023.06 | 10.18 | 28.98 | 39.82 | — | — | |
| LW-MCR2023.10 | 10.18 | 28.98 | 39.82 | — | — | |
| LW-MCR-uReference=TGRS’22, Backbone (image/text)=SqueezeNet, -, Params=1.65M2023.08 | 9.73 | 26.77 | 37.61 | 28.58 | — | |
| CLIP-BaselineParadigm=GVLM, Tuned on=None2023.06 | 9.51 | 23.01 | 32.74 | — | — | |
| CLIPBackbone=ViT-B/322026.07 | 9.51 | 23.01 | 32.96 | — | 145.4 | |
| Zero-shot CLIPReference=ICML’21, Backbone (image/text)=CLIP(ViT-B-32), Params=0.00M2023.08 | 9.29 | 26.33 | 37.39 | 23.89 | — | |
| SkyCLIPBackbone=ViT-B/322026.03 | 9.29 | 24.12 | 34.51 | — | — | |
| CAMP-bceReference=ICCV’19, Backbone (image/text)=Faster R-CNN, biGRU, Params=27.03M2023.08 | 9.07 | 23.01 | 33.19 | 22.03 | — | |
| LW-MCR-bReference=TGRS’22, Backbone (image/text)=SqueezeNet, -, Params=1.65M2023.08 | 9.07 | 22.79 | 38.05 | 25.55 | — | |
| VSE++Parms=29 M2023.10 | 9.07 | 21.61 | 31.78 | — | — | |
| SCAN i2tParms=60 M2023.10 | 8.92 | 22.12 | 33.78 | — | — | |
| LongCLIPBackbone=ViT-B/162026.03 | 8.85 | 23.67 | 32.96 | — | — | |
| CAMERAParms=64 M2023.10 | 8.33 | 21.83 | 33.11 | — | — | |
| CAMPParms=63 M2023.10 | 8.11 | 23.67 | 34.07 | — | — | |
| CLIPBackbone=ViT-B/162026.03 | 7.96 | 21.46 | 31.42 | — | — | |
| SCAN t2iParms=60 M2023.10 | 7.01 | 20.58 | 30.9 | — | — | |
| AMFMNParms=36 M2023.10 | — | — | — | 29.72 | — | |
| CAMERAParms=64 M2023.10 | — | — | — | 22.95 | — | |
| CAMPParms=63 M2023.10 | — | — | — | 23.47 | — | |
| DOVEParms=38 M2023.10 | — | — | — | 37.73 | — |