Image-to-text Retrieval on RSITMD
1RankRemoteCLIP
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| RemoteCLIPZero-shot protocol=True, Distribution setting=In-distribution2025.12 | 1 | 43.21 | 1 | — | — | — | |
| VLM2GeoVecZero-shot protocol=True, Distribution setting=Out-of-distribution2025.12 | 2 | 31.86 | 2.5 | — | — | — | |
| GeoRSCLIPZero-shot protocol=True, Distribution setting=Out-of-distribution2025.12 | 3 | 35.69 | 2.7 | — | — | — | |
| VLM2VecZero-shot protocol=True, Distribution setting=Out-of-distribution2025.12 | 4 | 29.27 | 4.2 | — | — | — | |
| CLIPZero-shot protocol=True, Distribution setting=Out-of-distribution2025.12 | 5 | 26.33 | 5 | — | — | — | |
| SkyCLIPZero-shot protocol=True, Distribution setting=Out-of-distribution2025.12 | 6 | 26.03 | 5.7 | — | — | — | |
| LW-MCRYear=2021, Pre-train data=02026.04 | 13.55 | — | — | 4.33 | 13.25 | 19.18 | |
| AMFMNYear=2022, Pre-train data=0, Lat. (ms)=150.1, Thr. (q/s)=6.72026.04 | 15.51 | — | — | 4.98 | 14.17 | 20.75 | |
| KAMCLYear=2023, Pre-train data=02026.04 | 21.38 | — | — | 6.42 | 18.69 | 36.58 | |
| CLIP-BaseYear=2021, Pre-train data=400M, large-scale VLM pre-training=true, Lat. (ms)=12.0, Thr. (q/s)=83.32026.04 | 22.81 | — | — | 9.11 | 20.42 | 31.7 | |
| CLIP-BaselineTraining Dataset=-, Zero-Shot=true2025.12 | 23.86 | — | — | 9.53 | 21.03 | 32.74 | |
| VSE++Training Dataset=RSITMD, Zero-Shot=false2025.12 | 24.54 | — | — | 10.28 | 27.65 | 39.6 | |
| LW-MCRTraining Dataset=RSITMD, Zero-Shot=false2025.12 | 27.79 | — | — | 10.18 | 28.98 | 39.82 | |
| AMFMNTraining Dataset=RSITMD, Zero-Shot=false2025.12 | 29.29 | — | — | 11.06 | 29.02 | 38.72 | |
| GaLRTraining Dataset=RSITMD, Zero-Shot=false2025.12 | 31.41 | — | — | 14.82 | 31.64 | 42.48 | |
| GaLRYear=2022, Pre-train data=0, Lat. (ms)=291.7, Thr. (q/s)=3.42026.04 | 31.41 | — | — | 11.15 | 36.68 | 51.68 | |
| SkyScriptYear=2024, Pre-train data=400M+2.6M, large-scale VLM pre-training=true, Lat. (ms)=13.5, Thr. (q/s)=74.12026.04 | 31.48 | — | — | 12.05 | 35.74 | 53.68 | |
| HVSATraining Dataset=RSITMD, Zero-Shot=false2025.12 | 33.15 | — | — | 13.2 | 32.08 | 45.58 | |
| SWANTraining Dataset=RSITMD, Zero-Shot=false2025.12 | 33.89 | — | — | 13.35 | 32.55 | 45.11 | |
| GeoRSCLIPTraining Dataset=RS5M, Zero-Shot=false2025.12 | 35.68 | — | — | 19.03 | 34.51 | 46.46 | |
| FAAMIYear=2023, Pre-train data=02026.04 | 35.99 | — | — | 16.15 | 35.62 | 48.89 | |
| VGSGNYear=2024, Pre-train data=0, Lat. (ms)=106.5, Thr. (q/s)=9.42026.04 | 36.51 | — | — | 14.16 | 34.96 | 50.66 | |
| FAAMITraining Dataset=RSITMD, Zero-Shot=false2025.12 | 36.52 | — | — | 19.32 | 35.62 | 48.89 | |
| PIRYear=2023, Pre-train data=0, Lat. (ms)=117.6, Thr. (q/s)=8.52026.04 | 38.24 | — | — | 18.14 | 41.15 | 52.88 | |
| PIRTraining Dataset=RSITMD, Zero-Shot=false2025.12 | 38.41 | — | — | 17.64 | 41.15 | 53.82 | |
| MTGFEYear=2023, Pre-train data=02026.04 | 40.38 | — | — | 16.33 | 48.15 | 66.85 | |
| MTGFETraining Dataset=RSITMD, Zero-Shot=false2025.12 | 40.78 | — | — | 17.92 | 40.93 | 53.32 | |
| MultilanguageTraining Dataset=RSITMD, Zero-Shot=false2025.12 | 41.38 | — | — | 19.69 | 40.26 | 54.42 | |
| ML RetrievalTraining Dataset=RSITMD, Backbone=ViT-B/322026.03 | 41.38 | — | — | 19.09 | 40.26 | 54.42 | |
| MSCMATYear=2024, Pre-train data=02026.04 | 41.38 | — | — | 22.35 | 42.92 | 55.75 | |
| TRSLLaVATraining Dataset=-, Zero-Shot=true2025.12 | 42.62 | — | — | 19.54 | 44.91 | 61.28 | |
| SIPNYear=2025, Pre-train data=0, Lat. (ms)=120.3, Thr. (q/s)=8.32026.04 | 43.47 | — | — | 23.01 | 46.02 | 57.96 | |
| GRAFTYear=2024, Pre-train data=10.2M, large-scale VLM pre-training=true2026.04 | 44.61 | — | — | 25.25 | 44.68 | 57.89 | |
| PETLYear=2023, Pre-train data=400M, large-scale VLM pre-training=true2026.04 | 45.33 | — | — | 24.16 | 47.12 | 61.28 | |
| CLIPTraining Dataset=RET-3, Backbone=ViT-B/322026.03 | 47.63 | — | — | 23.67 | 50 | 63.27 | |
| IERRYear=2024, Pre-train data=02026.04 | 48.55 | — | — | 27.66 | 48.89 | 61.06 | |
| FTFYear=2026, Pre-train data=0, Candidate set size (K)=50, Lat. (ms)=24.7, Thr. (q/s)=40.62026.04 | 48.65 | — | — | 27.8 | 49.3 | 63.4 | |
| RemoteCLIPTraining Dataset=RET-3 + DET-10 + SEG-4, Backbone=ViT-B/322026.03 | 49.38 | — | — | 27.88 | 50.66 | 65.71 | |
| RemoteCLIPYear=2024, Pre-train data=400M+, large-scale VLM pre-training=true, Lat. (ms)=12.5, Thr. (q/s)=80.02026.04 | 49.38 | — | — | 27.88 | 50.66 | 65.71 | |
| EAPAYear=2025, Pre-train data=400M, large-scale VLM pre-training=true2026.04 | 49.91 | — | — | 29.87 | 53.32 | 62.83 | |
| GeoRSCLIPYear=2024, Pre-train data=400M+5M, large-scale VLM pre-training=true, Lat. (ms)=13.0, Thr. (q/s)=76.92026.04 | 50.1 | — | — | 30.09 | 51.55 | 63.27 | |
| FTFYear=2026, Pre-train data=0, Candidate set size (K)=100, Lat. (ms)=47.7, Thr. (q/s)=21.02026.04 | 50.49 | — | — | 29.6 | 51.2 | 65.3 | |
| FSSNYear=2025, Pre-train data=0, Lat. (ms)=128.7, Thr. (q/s)=7.82026.04 | 51.28 | — | — | 30.53 | 52.21 | 65.49 | |
| FTFYear=2026, Pre-train data=0, Candidate set size (K)=200, Lat. (ms)=93.7, Thr. (q/s)=10.72026.04 | 51.47 | — | — | 30.3 | 52.4 | 66.2 | |
| GeoRSCLIP-FTTraining Dataset=RSITMD + RSICD, Backbone=ViT-B/32, Evaluation Protocol=full fine-tuning2026.03 | 51.81 | — | — | 32.3 | 53.32 | 67.92 | |
| AVIONTraining Dataset=RSITMD + RSICD, Backbone=ViT-B/32, Evaluation Protocol=lightweight prompt learning2026.03 | 52.92 | — | — | 33.17 | 54.46 | 69.18 | |
| EBAKERYear=2024, Pre-train data=400M+0.2M, large-scale VLM pre-training=true, Lat. (ms)=13.6, Thr. (q/s)=73.92026.04 | 53.32 | — | — | 34.07 | 54.2 | 67.95 | |
| Geo-RSClip+RS5Mzero-shot=true2025.10 | — | 33.33 | — | — | — | — | |
| MaMMUT 400Mzero-shot=true, parameters=400M2025.10 | — | 28.83 | — | — | — | — | |
| PIR-ITRzero-shot=true2025.10 | — | 38.64 | — | — | — | — | |
| RS-MaMMut 400Mzero-shot=true, parameters=400M2025.10 | — | 42.63 | — | — | — | — | |
| RS-SigLIP2 400Mzero-shot=true, parameters=400M2025.10 | — | 43.14 | — | — | — | — | |
| SigLIP2 400Mzero-shot=true, parameters=400M2025.10 | — | 24.62 | — | — | — | — | |
| SkyScript SkyCLIP-30zero-shot=true2025.10 | — | 30.75 | — | — | — | — |