List your company

Text-to-Image Retrieval on MS-COCO

67R@1BLIP (Cosine)
34.271242.768151.26559.7619Apr 1, 2020Mar 14, 2021Feb 25, 2022Feb 7, 2023Jan 21, 2024Jan 2, 2025Dec 16, 2025

Evaluation results

MethodLinks
BLIP (Cosine)Backbone=BLIP, Alignment Strategy=Cosine2025.066791.295.9———
L2RM-SGRAF2024.0365.790.896.1528.6——
GSC-SGRNoise=20%2024.0564.490.695.9525.7——
RCL-SGRAF2024.0364.390.896526.6——
MSCN-SGRNoise=20%2024.0564.390.495.8524.6——
L2RM-SGR2024.0363.790.295.8523.4——
BiCro-SGRAFNoise=20%2024.0563.790.395.7523.2——
L2RM-SAF2024.0363.489.695.1520.6——
GSMN (sparse+dense)Image Backbone=Faster R-CNN, Text Backbone=ft Bi-GRU2020.0463.390.195.7522.5——
NCR2024.0363.390.495.8522.5——
DECL-SGRAF2024.0363.390.695523.4——
SGRAF2024.0363.290.796.1524.3——
RCL-SGR2024.0362.99095.7521.4——
RCL-SAF2024.0362.79095.4521.3——
BiCro2024.0362.689.795.7520.5——
GSC-SGRNoise=40%2024.0562.589.795.4519.9——
SAF2024.0361.889.495.3516.3——
BiCro-SGRAFNoise=40%2024.0561.889.294.9517.1——
DECL-SGRAFNoise=20%2024.0561.789.395.4518.2——
RINCE-SGRNoise=20%2024.0561.789.294.7513.5——
PFANImage Backbone=Faster R-CNN, Text Backbone=ft Bi-GRU2020.0461.689.695.2518.2——
SGR2024.0361.489.395.4518.1——
NCR-SGRNoise=20%2024.0560.888.895515——
IMRAM2024.0360.688.994.6512.1——
GSMN (sparse)Image Backbone=Faster R-CNN, Text Backbone=ft Bi-GRU2020.0460.488.795514——
GSMN (dense)Image Backbone=Faster R-CNN, Text Backbone=ft Bi-GRU2020.0460.388.594.6511.6——
MSCN-SGRNoise=40%2024.0560.388.694.8512.6——
SGRAFNoise=20%2024.0560.188.594.8511.9——
GSC-SGRNoise=60%2024.056088.394.6511.7——
NCR-SGRNoise=40%2024.0559.688.194.7509.7——
DECL-SGRAFNoise=40%2024.0559.588.394.8512——
BFANImage Backbone=Faster R-CNN, Text Backbone=ft Bi-GRU2020.0459.488.4—317.9——
RINCE-SGRNoise=40%2024.0559.188.694.3506.9——
SCANImage Backbone=Faster R-CNN, Text Backbone=ft Bi-GRU2020.0458.888.494.8507.9——
RINCE-SGRNoise=60%2024.0558.486.692.5501.7——
BiCro-SGRAFNoise=60%2024.0558.387.293.9505.5——
MSCN-SGRNoise=60%2024.0557.78793.9501.8——
DECL-SGRAFNoise=60%2024.055786.693.8502.5——
SCOImage Backbone=fixed ResNet-152, Text Backbone=ft LSTM2020.0456.787.594.8499.3——
GXNImage Backbone=ft ResNet-152, Text Backbone=ft GRU2020.0456.6—94.5317.5——
FIMA-Q w/ RegCacheBackbone=SigLIP2-B/16, Precision=W8A8, Evaluation Protocol=Zero-shot2025.1052.5—————
FIMA-Q w/ RegCacheBackbone=SigLIP2-B/16, Precision=W4A4, Evaluation Protocol=Zero-shot2025.1052.49—————
FIMA-Q w/ RegCacheBackbone=SigLIP2-B/16, Precision=W6A6, Evaluation Protocol=Zero-shot2025.1052.49—————
CARTParadigm=Coarse-Fine ID2024.0652.4277.5386.11———
FP32Backbone=SigLIP2-B/16, Precision=FP32, Evaluation Protocol=Zero-shot2025.1052.33—————
VSE++Image Backbone=ft ResNet-152, Text Backbone=ft GRU2020.0452—92304.6——
PTQ4ViT w/ RegCacheBackbone=SigLIP2-B/16, Precision=W8A8, Evaluation Protocol=Zero-shot2025.1051.68—————
PTQ4ViTBackbone=SigLIP2-B/16, Precision=W8A8, Evaluation Protocol=Zero-shot2025.1051.62—————
NoisyQuant w/ RegCacheBackbone=SigLIP2-B/16, Precision=W8A8, Evaluation Protocol=Zero-shot2025.1051.41—————
MobileCLIP-BEvaluation Protocol=zero-shot2023.1150.674.982.9———
ERQ w/ RegCacheBackbone=SigLIP2-B/16, Precision=W8A8, Evaluation Protocol=Zero-shot2025.1050.2—————
RepQ-ViTBackbone=SigLIP2-B/16, Precision=W8A8, Evaluation Protocol=Zero-shot2025.1050.15—————
RepQ-ViT w/ RegCacheBackbone=SigLIP2-B/16, Precision=W8A8, Evaluation Protocol=Zero-shot2025.1050.15—————
ERQBackbone=SigLIP2-B/16, Precision=W8A8, Evaluation Protocol=Zero-shot2025.1049.97—————
CLIP (Cosine)Backbone=CLIP, Alignment Strategy=Cosine2025.0649.376.488.2———
Fix-CLIP2025.1149.173.882.4———
ViCLIP-B/16Evaluation Protocol=zero-shot2023.1148.473.381.8———
FIMA-QBackbone=SigLIP2-B/16, Precision=W4A4, Evaluation Protocol=Zero-shot2025.1047.22—————
FIMA-QBackbone=SigLIP2-B/16, Precision=W8A8, Evaluation Protocol=Zero-shot2025.1047.22—————
FIMA-QBackbone=SigLIP2-B/16, Precision=W6A6, Evaluation Protocol=Zero-shot2025.1047.21—————
DPCImage Backbone=ft ResNet-152, Text Backbone=ft ResNet-1522020.0447.179.990467.8——
Naïve w/ RegCacheBackbone=SigLIP2-B/16, Precision=W8A8, Evaluation Protocol=Zero-shot2025.1046.8—————
CLIP-MoE2025.1146.871.780.4———
CLIP-ERMoE2025.1146.572.981.6———
NoisyQuantBackbone=SigLIP2-B/16, Precision=W8A8, Evaluation Protocol=Zero-shot2025.1046.32—————
Long-CLIP2025.1146.370.879.8———
ERQ w/ RegCacheBackbone=SigLIP2-B/16, Precision=W6A6, Evaluation Protocol=Zero-shot2025.1046.28—————
ERQBackbone=SigLIP2-B/16, Precision=W6A6, Evaluation Protocol=Zero-shot2025.1045.44—————
MobileCLIP-S2Evaluation Protocol=zero-shot2023.1145.470.179———
MobileCLIP-S1Evaluation Protocol=zero-shot2023.114468.977.7———
CLIPMode=Zero-shot2025.1242.266.9————
OursMode=Zero-shot2025.1242.167.3————
B-cosifiedMode=Zero-shot2025.1241.966.7————
GRACEParadigm=Atomic ID2024.0641.569.179.1———
FP32Backbone=OpenCLIP-B/16, Precision=FP32, Evaluation Protocol=Zero-shot2025.1041.38—————
RepQ-ViT w/ RegCacheBackbone=SigLIP2-B/16, Precision=W6A6, Evaluation Protocol=Zero-shot2025.1041.26—————
PTQ4ViT w/ RegCacheBackbone=OpenCLIP-B/16, Precision=W6A6, Evaluation Protocol=Zero-shot2025.1040.96—————
PTQ4ViT w/ RegCacheBackbone=OpenCLIP-B/16, Precision=W8A8, Evaluation Protocol=Zero-shot2025.1040.96—————
PTQ4ViTBackbone=OpenCLIP-B/16, Precision=W8A8, Evaluation Protocol=Zero-shot2025.1040.66—————
FIMA-Q w/ RegCacheBackbone=OpenCLIP-B/16, Precision=W8A8, Evaluation Protocol=Zero-shot2025.1040.58—————
MobileCLIP-S0Evaluation Protocol=zero-shot2023.1140.46675.9———
FIMA-QBackbone=OpenCLIP-B/16, Precision=W6A6, Evaluation Protocol=Zero-shot2025.1040.39—————
FIMA-Q w/ RegCacheBackbone=OpenCLIP-B/16, Precision=W6A6, Evaluation Protocol=Zero-shot2025.1040.3—————
FIMA-QBackbone=OpenCLIP-B/16, Precision=W8A8, Evaluation Protocol=Zero-shot2025.1040.21—————
ERQ w/ RegCacheBackbone=OpenCLIP-B/16, Precision=W8A8, Evaluation Protocol=Zero-shot2025.1040.04—————
FIMA-Q w/ RegCacheBackbone=OpenCLIP-B/16, Precision=W4A4, Evaluation Protocol=Zero-shot2025.1040.02—————
PTQ4ViT w/ RegCacheBackbone=SigLIP2-B/16, Precision=W6A6, Evaluation Protocol=Zero-shot2025.1039.96—————
ERQBackbone=OpenCLIP-B/16, Precision=W8A8, Evaluation Protocol=Zero-shot2025.1039.83—————
RepQ-ViT w/ RegCacheBackbone=OpenCLIP-B/16, Precision=W8A8, Evaluation Protocol=Zero-shot2025.1039.82—————
DSPEImage Backbone=fixed VGG-19, Text Backbone=w2v+HGLMM2020.0439.675.286.9420.7——
NoisyQuant w/ RegCacheBackbone=SigLIP2-B/16, Precision=W6A6, Evaluation Protocol=Zero-shot2025.1039.25—————
RepQ-ViTBackbone=SigLIP2-B/16, Precision=W6A6, Evaluation Protocol=Zero-shot2025.1039.2—————
RepQ-ViTBackbone=OpenCLIP-B/16, Precision=W8A8, Evaluation Protocol=Zero-shot2025.1038.88—————
CLIP (Manhattan)Backbone=CLIP, Alignment Strategy=Manhattan2025.0638.868.780.1———
TinyCLIP-63M/32Evaluation Protocol=zero-shot2023.1138.5—————
Naïve w/ RegCacheBackbone=OpenCLIP-B/16, Precision=W8A8, Evaluation Protocol=Zero-shot2025.1038.45—————
BLIP (Transformer MSE)Backbone=BLIP, Alignment Strategy=Transformer (MSE)2025.0638.470.281.9———
FIMA-QBackbone=OpenCLIP-B/16, Precision=W4A4, Evaluation Protocol=Zero-shot2025.1038.02—————
Ours-RGB0.3strategy=RGB, anchor_patch_ratio=0.3, mode=Zero-shot2024.0535.8761.572.34———
NoisyQuant w/ RegCacheBackbone=OpenCLIP-B/16, Precision=W8A8, Evaluation Protocol=Zero-shot2025.1035.53—————