Image-Text Retrieval on RSICD
40.72Mean RecallGeoMELT
Evaluation results
| Method | Links | |||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| GeoMELTVision Backbone=MWT, Finetuning Data=Multi-Task Data, Total Params (M)=2712025.12 | 40.72 | 22.87 | 44.1 | 56.63 | 17.2 | 43.59 | 59.93 | — | — | — | — | — | — | |
| AVIONTraining Dataset=RSITMD + RSICD, Backbone=ViT-B/32, Evaluation Protocol=lightweight prompt learning2026.03 | 39.8 | — | — | — | — | — | — | 22.01 | 42.37 | 56.42 | — | — | — | |
| GeoRSCLIPVision Backbone=ViT-B, Finetuning Data=RS5M + RSICD + RSITMD, Total Params (M)=1512025.12 | 38.87 | 15.59 | 41.19 | 57.99 | 21.13 | 41.72 | 55.63 | — | — | — | — | — | — | |
| GeoRSCLIP-FTTraining Dataset=RSITMD + RSICD, Backbone=ViT-B/32, Evaluation Protocol=full fine-tuning2026.03 | 38.87 | — | — | — | — | — | — | 21.13 | 41.72 | 55.63 | — | — | — | |
| RemoteCLIP (ViT-L)Vision Backbone=ViT-L, Finetuning Data=RET-3 + DET-10 + SEG-4, Total Params (M)=4282025.12 | 36.35 | 18.39 | 37.42 | 51.05 | 14.73 | 39.93 | 56.58 | — | — | — | — | — | — | |
| RemoteCLIP (ViT-B)Vision Backbone=ViT-B, Finetuning Data=RET-3 + DET-10 + SEG-4, Total Params (M)=1512025.12 | 35.26 | 17.02 | 37.97 | 51.51 | 13.71 | 37.11 | 54.25 | — | — | — | — | — | — | |
| RemoteCLIPTraining Dataset=RET-3 + DET-10 + SEG-4, Backbone=ViT-B/322026.03 | 35.26 | — | — | — | — | — | — | 17.02 | 39.97 | 51.51 | — | — | — | |
| CLIPTraining Dataset=RET-3, Backbone=ViT-B/322026.03 | 33.98 | — | — | — | — | — | — | 17.94 | 35.94 | 51.04 | — | — | — | |
| RRSITRNoise=20%2026.03 | 32.7 | — | — | — | — | — | — | 16.56 | 34.8 | 48.01 | — | — | — | |
| RRSITRNoise=20%2026.03 | 32.7 | — | — | — | — | — | — | 12.5 | 34.22 | 50.12 | — | — | — | |
| RRSITRNoise=40%2026.03 | 32.51 | — | — | — | — | — | — | 15.77 | 34.15 | 47.36 | — | — | — | |
| RRSITRNoise=40%2026.03 | 32.51 | — | — | — | — | — | — | 11.81 | 35.09 | 50.9 | — | — | — | |
| RRSITRNoise=60%2026.03 | 31.98 | — | — | — | — | — | — | 15.75 | 33.91 | 47.28 | — | — | — | |
| RRSITRNoise=60%2026.03 | 31.98 | — | — | — | — | — | — | 11.92 | 33.81 | 49.23 | — | — | — | |
| TRSLLaVAZero-Shot=true2025.12 | 31.33 | — | — | — | — | — | — | 9.38 | 28.35 | 44.01 | — | — | — | |
| MTGFETraining Dataset=RSICD, Zero-Shot=false2025.12 | 30.68 | — | — | — | — | — | — | 8.67 | 27.56 | 43.92 | — | — | — | |
| CLIP-RSICDVision Backbone=ViT-B, Finetuning Data=RSICD, Total Params (M)=1512025.12 | 30.19 | 14.09 | 30.1 | 43.64 | 11.16 | 33.25 | 48.91 | — | — | — | — | — | — | |
| RRSITRNoise=80%2026.03 | 28.23 | — | — | — | — | — | — | 13.47 | 28.8 | 41.41 | — | — | — | |
| RRSITRNoise=80%2026.03 | 28.23 | — | — | — | — | — | — | 10.02 | 30.55 | 45.13 | — | — | — | |
| Rahhal et al.Vision Backbone=ViT-B, Finetuning Data=RSICD, Total Params (M)=1512025.12 | 27.43 | 10.7 | 29.64 | 41.53 | 9.14 | 28.96 | 44.59 | — | — | — | — | — | — | |
| ML RetrievalTraining Dataset=RSICD, Backbone=ViT-B/322026.03 | 27.43 | — | — | — | — | — | — | 10.7 | 29.64 | 41.53 | — | — | — | |
| PIRTraining Dataset=RSICD, Zero-Shot=false2025.12 | 27.16 | — | — | — | — | — | — | 9.14 | 28.96 | 44.59 | — | — | — | |
| CUPNoise=20%2026.03 | 26.56 | — | — | — | — | — | — | 9.51 | 29.29 | 44.01 | — | — | — | |
| CUPNoise=20%2026.03 | 26.56 | — | — | — | — | — | — | 7.45 | 26.82 | 42.29 | — | — | — | |
| GeoRSCLIPTraining Dataset=RS5M, Zero-Shot=false2025.12 | 25.69 | — | — | — | — | — | — | 9.52 | 27.37 | 40.99 | — | — | — | |
| CUPNoise=40%2026.03 | 25.2 | — | — | — | — | — | — | 9.11 | 28.16 | 42.16 | — | — | — | |
| CUPNoise=40%2026.03 | 25.2 | — | — | — | — | — | — | 7.33 | 24.83 | 39.63 | — | — | — | |
| FAAMITraining Dataset=RSICD, Zero-Shot=false2025.12 | 23.18 | — | — | — | — | — | — | 8.11 | 25.59 | 41.37 | — | — | — | |
| MultilanguageTraining Dataset=RSICD, Zero-Shot=false2025.12 | 23.18 | — | — | — | — | — | — | 8.11 | 25.59 | 41.37 | — | — | — | |
| CUPNoise=60%2026.03 | 22.58 | — | — | — | — | — | — | 7.21 | 24.58 | 38.61 | — | — | — | |
| CUPNoise=60%2026.03 | 22.58 | — | — | — | — | — | — | 6.21 | 22.51 | 36.34 | — | — | — | |
| MSANoise=20%2026.03 | 20.62 | — | — | — | — | — | — | 6.99 | 20.28 | 31.77 | — | — | — | |
| MSANoise=20%2026.03 | 20.62 | — | — | — | — | — | — | 5.65 | 21.93 | 37.12 | — | — | — | |
| SWANTraining Dataset=RSICD, Zero-Shot=false2025.12 | 20.61 | — | — | — | — | — | — | 5.56 | 22.26 | 37.41 | — | — | — | |
| KAMCLNoise=20%2026.03 | 20.2 | — | — | — | — | — | — | 7.59 | 20.48 | 31.31 | — | — | — | |
| KAMCLNoise=20%2026.03 | 20.2 | — | — | — | — | — | — | 5.75 | 21.38 | 34.7 | — | — | — | |
| HVSATraining Dataset=RSICD, Zero-Shot=false2025.12 | 20.16 | — | — | — | — | — | — | 5.51 | 21.13 | 34.13 | — | — | — | |
| PIRNoise=20%2026.03 | 20.04 | — | — | — | — | — | — | 7.36 | 20.29 | 31.64 | — | — | — | |
| PIRNoise=20%2026.03 | 20.04 | — | — | — | — | — | — | 5.41 | 20.68 | 34.84 | — | — | — | |
| CUPNoise=80%2026.03 | 19.82 | — | — | — | — | — | — | 5.87 | 20.93 | 34.33 | — | — | — | |
| CUPNoise=80%2026.03 | 19.82 | — | — | — | — | — | — | 5.51 | 19.7 | 32.59 | — | — | — | |
| KCRVision Backbone=ResNet18, Finetuning Data=RSICD, Total Params (M)=11*2025.12 | 19.14 | 4.76 | 18.59 | 27.2 | 5.84 | 22.31 | 36.12 | — | — | — | — | — | — | |
| GaLRVision Backbone=ResNet18, Finetuning Data=RSICD, Total Params (M)=11*2025.12 | 18.96 | 6.59 | 19.85 | 31.04 | 4.69 | 19.48 | 32.13 | — | — | — | — | — | — | |
| GaLRTraining Dataset=RSICD, Zero-Shot=false2025.12 | 18.96 | — | — | — | — | — | — | 4.69 | 19.48 | 32.13 | — | — | — | |
| CMFM-NetVision Backbone=ResNet18, Finetuning Data=RSICD, Total Params (M)=11*2025.12 | 17.75 | 5.4 | 18.66 | 28.55 | 5.31 | 18.57 | 30.03 | — | — | — | — | — | — | |
| CLIP (ViT-L)Vision Backbone=ViT-L, Finetuning Data=Zero-shot, Total Params (M)=4282025.12 | 17.56 | 6.04 | 17.48 | 27.54 | 5.03 | 19.03 | 30.25 | — | — | — | — | — | — | |
| KCRTraining Dataset=RSICD, Zero-Shot=false2025.12 | 17.35 | — | — | — | — | — | — | 4.76 | 18.59 | 27.1 | — | — | — | |
| SEMICLIPNoise=20%2026.03 | 16.94 | — | — | — | — | — | — | 4.85 | 17.59 | 29.24 | — | — | — | |
| SEMICLIPNoise=20%2026.03 | 16.94 | — | — | — | — | — | — | 4.28 | 17.13 | 28.54 | — | — | — | |
| AMFMNVision Backbone=ResNet50, Finetuning Data=RSICD, Total Params (M)=25*2025.12 | 16.42 | 5.39 | 15.08 | 23.4 | 4.9 | 18.28 | 31.44 | — | — | — | — | — | — | |
| AMFMNTraining Dataset=RSICD, Zero-Shot=false2025.12 | 16.42 | — | — | — | — | — | — | 4.9 | 18.28 | 31.44 | — | — | — | |
| PIRNoise=40%2026.03 | 15.68 | — | — | — | — | — | — | 4.61 | 14.99 | 24.67 | — | — | — | |
| PIRNoise=40%2026.03 | 15.68 | — | — | — | — | — | — | 3.8 | 16.48 | 29.53 | — | — | — | |
| CLIP-BaselineZero-Shot=true2025.12 | 15.64 | — | — | — | — | — | — | 5.78 | 17.73 | 27.16 | — | — | — | |
| CLIP (ViT-B)Vision Backbone=ViT-B, Finetuning Data=Zero-shot, Total Params (M)=1512025.12 | 15.62 | 4.58 | 14.55 | 23.7 | 5.8 | 16.85 | 28.23 | — | — | — | — | — | — | |
| AMFMNNoise=20%2026.03 | 15.47 | — | — | — | — | — | — | 4.68 | 15.08 | 24.99 | — | — | — | |
| AMFMNNoise=20%2026.03 | 15.47 | — | — | — | — | — | — | 4.08 | 16.31 | 27.65 | — | — | — | |
| SIRSNoise=20%2026.03 | 15.39 | — | — | — | — | — | — | 4.8 | 14.38 | 24.17 | — | — | — | |
| SIRSNoise=20%2026.03 | 15.39 | — | — | — | — | — | — | 3.48 | 16.11 | 29.41 | — | — | — | |
| S-CLIPNoise=20%2026.03 | 15.19 | — | — | — | — | — | — | 3.82 | 14.84 | 26.26 | — | — | — | |
| S-CLIPNoise=20%2026.03 | 15.19 | — | — | — | — | — | — | 3.9 | 15.13 | 27.19 | — | — | — | |
| SWANNoise=20%2026.03 | 14.73 | — | — | — | — | — | — | 4.5 | 13.76 | 22.71 | — | — | — | |
| SWANNoise=20%2026.03 | 14.73 | — | — | — | — | — | — | 3.45 | 15.43 | 28.54 | — | — | — | |
| LW-MCRTraining Dataset=RSICD, Zero-Shot=false2025.12 | 14.66 | — | — | — | — | — | — | 4.66 | 17.51 | 30.02 | — | — | — | |
| HVSANoise=20%2026.03 | 14.29 | — | — | — | — | — | — | 4.76 | 14.93 | 25.21 | — | — | — | |
| HVSANoise=20%2026.03 | 14.29 | — | — | — | — | — | — | 2.85 | 13.15 | 24.85 | — | — | — | |
| AMFMNNoise=40%2026.03 | 13.01 | — | — | — | — | — | — | 3.68 | 13.03 | 21.67 | — | — | — | |
| AMFMNNoise=40%2026.03 | 13.01 | — | — | — | — | — | — | 3.09 | 12.95 | 23.62 | — | — | — | |
| SWANNoise=40%2026.03 | 12.59 | — | — | — | — | — | — | 3.31 | 10.67 | 19.87 | — | — | — | |
| SWANNoise=40%2026.03 | 12.59 | — | — | — | — | — | — | 2.97 | 13.23 | 25.49 | — | — | — | |
| SEMICLIPNoise=40%2026.03 | 12.56 | — | — | — | — | — | — | 3.66 | 13.7 | 22.58 | — | — | — | |
| SEMICLIPNoise=40%2026.03 | 12.56 | — | — | — | — | — | — | 2.91 | 11.91 | 20.62 | — | — | — | |
| SIRSNoise=40%2026.03 | 12.52 | — | — | — | — | — | — | 3.28 | 11.36 | 20.38 | — | — | — | |
| SIRSNoise=40%2026.03 | 12.52 | — | — | — | — | — | — | 2.9 | 13.14 | 24.07 | — | — | — | |
| S-CLIPNoise=40%2026.03 | 11.93 | — | — | — | — | — | — | 3.09 | 12.31 | 21.17 | — | — | — | |
| S-CLIPNoise=40%2026.03 | 11.93 | — | — | — | — | — | — | 2.85 | 11.47 | 20.7 | — | — | — | |
| PIRNoise=60%2026.03 | 11.85 | — | — | — | — | — | — | 3.35 | 10.65 | 18.63 | — | — | — | |
| PIRNoise=60%2026.03 | 11.85 | — | — | — | — | — | — | 2.79 | 12.6 | 23.06 | — | — | — | |
| HVSANoise=40%2026.03 | 11.49 | — | — | — | — | — | — | 3.47 | 13.32 | 22.16 | — | — | — | |
| HVSANoise=40%2026.03 | 11.49 | — | — | — | — | — | — | 1.82 | 9.84 | 18.34 | — | — | — | |
| SWANNoise=60%2026.03 | 11.34 | — | — | — | — | — | — | 2.8 | 8.84 | 16.72 | — | — | — | |
| SWANNoise=60%2026.03 | 11.34 | — | — | — | — | — | — | 2.82 | 12.88 | 23.95 | — | — | — | |
| MSANoise=40%2026.03 | 11.16 | — | — | — | — | — | — | 2.19 | 7.63 | 13.83 | — | — | — | |
| MSANoise=40%2026.03 | 11.16 | — | — | — | — | — | — | 3.27 | 14.04 | 26 | — | — | — | |
| KAMCLNoise=40%2026.03 | 11.06 | — | — | — | — | — | — | 3.49 | 10.5 | 17.8 | — | — | — | |
| KAMCLNoise=40%2026.03 | 11.06 | — | — | — | — | — | — | 2.41 | 11.11 | 21.06 | — | — | — | |
| AMFMNNoise=60%2026.03 | 10.95 | — | — | — | — | — | — | 3.44 | 10.87 | 18.87 | — | — | — | |
| AMFMNNoise=60%2026.03 | 10.95 | — | — | — | — | — | — | 2.49 | 10.75 | 19.29 | — | — | — | |
| PIRNoise=80%2026.03 | 10.64 | — | — | — | — | — | — | 2.49 | 9.57 | 18.41 | — | — | — | |
| PIRNoise=80%2026.03 | 10.64 | — | — | — | — | — | — | 2.1 | 10.8 | 20.46 | — | — | — | |
| VSE++Training Dataset=RSICD, Zero-Shot=false2025.12 | 10.43 | — | — | — | — | — | — | 2.82 | 11.32 | 18.1 | — | — | — | |
| HVSANoise=60%2026.03 | 10.4 | — | — | — | — | — | — | 3.5 | 12.42 | 20.88 | — | — | — | |
| HVSANoise=60%2026.03 | 10.4 | — | — | — | — | — | — | 1.72 | 8.37 | 15.49 | — | — | — | |
| AMFMNNoise=80%2026.03 | 9.06 | — | — | — | — | — | — | 2.51 | 8.97 | 16.27 | — | — | — | |
| AMFMNNoise=80%2026.03 | 9.06 | — | — | — | — | — | — | 1.88 | 8.67 | 16.08 | — | — | — | |
| SEMICLIPNoise=60%2026.03 | 8.87 | — | — | — | — | — | — | 2.34 | 9.28 | 16.14 | — | — | — | |
| SEMICLIPNoise=60%2026.03 | 8.87 | — | — | — | — | — | — | 1.7 | 8.69 | 15.09 | — | — | — | |
| SIRSNoise=60%2026.03 | 8.01 | — | — | — | — | — | — | 2.32 | 7.1 | 13.63 | — | — | — | |
| SIRSNoise=60%2026.03 | 8.01 | — | — | — | — | — | — | 1.65 | 8.03 | 15.31 | — | — | — | |
| SWANNoise=80%2026.03 | 7.44 | — | — | — | — | — | — | 1.74 | 6 | 11.6 | — | — | — |