List your company

Image-to-Text Retrieval on MS-COCO

80.5R@1L2RM-SGRAF
20.28435.91751.5567.183Nov 28, 2023Mar 31, 2024Aug 3, 2024Dec 6, 2024Apr 10, 2025Aug 13, 2025Dec 16, 2025

Evaluation results

MethodLinks
L2RM-SGRAF2024.0380.596.698.9528.6—
RCL-SGRAF2024.0380.496.498.7526.6—
SGRAF2024.0379.696.298.5524.3—
GSC-SGRNoise=20%2024.0579.596.498.9525.7—
DECL-SGRAF2024.0379.396.598.7523.4—
L2RM-SGR2024.037996.498.3523.4—
BiCro-SGRAFNoise=20%2024.0578.896.198.6523.2—
NCR2024.0378.795.898.5522.5—
RCL-SAF2024.0378.596.198.6521.3—
BiCro2024.0378.495.698.5520.5—
RCL-SGR2024.0378.296.298.4521.4—
L2RM-SAF2024.0378.295.798.6520.6—
GSC-SGRNoise=40%2024.0578.295.998.2519.9—
MSCN-SGRNoise=20%2024.0578.197.298.8524.6—
SGR2024.037895.898.2518.1—
DECL-SGRAFNoise=20%2024.0577.595.998.4518.2—
BiCro-SGRAFNoise=40%2024.057795.998.3517.1—
NCR-SGRNoise=20%2024.0576.695.698.2515—
SAF2024.0376.195.498.3516.3—
DECL-SGRAFNoise=40%2024.0575.695.598.3512—
GSC-SGRNoise=60%2024.0575.695.198511.7—
SGRAFNoise=20%2024.0575.495.297.9511.9—
MSCN-SGRNoise=40%2024.0575.395.498.2512.6—
NCR-SGRNoise=40%2024.0574.794.698509.7—
IMRAM2024.037495.698.4512.1—
BiCro-SGRAFNoise=60%2024.0573.994.497.8505.5—
RINCE-SGRNoise=20%2024.0573.895.698.5513.5—
DECL-SGRAFNoise=60%2024.057394.297.9502.5—
MSCN-SGRNoise=60%2024.0572.593.697.1501.8—
RINCE-SGRNoise=60%2024.0572.39497.9501.7—
FP32Backbone=SigLIP2-B/16, Precision=FP32, Evaluation Protocol=Zero-shot2025.1071.6————
FIMA-Q w/ RegCacheBackbone=SigLIP2-B/16, Precision=W4A4, Evaluation Protocol=Zero-shot2025.1071.54————
FIMA-Q w/ RegCacheBackbone=SigLIP2-B/16, Precision=W6A6, Evaluation Protocol=Zero-shot2025.1071.54————
FIMA-Q w/ RegCacheBackbone=SigLIP2-B/16, Precision=W8A8, Evaluation Protocol=Zero-shot2025.1071.52————
RINCE-SGRNoise=40%2024.0571.295.897.9506.9—
NoisyQuant w/ RegCacheBackbone=SigLIP2-B/16, Precision=W8A8, Evaluation Protocol=Zero-shot2025.1070.24————
PTQ4ViT w/ RegCacheBackbone=SigLIP2-B/16, Precision=W8A8, Evaluation Protocol=Zero-shot2025.1070.12————
ERQ w/ RegCacheBackbone=SigLIP2-B/16, Precision=W8A8, Evaluation Protocol=Zero-shot2025.1070————
RepQ-ViT w/ RegCacheBackbone=SigLIP2-B/16, Precision=W8A8, Evaluation Protocol=Zero-shot2025.1069.76————
ERQBackbone=SigLIP2-B/16, Precision=W8A8, Evaluation Protocol=Zero-shot2025.1069.76————
PTQ4ViTBackbone=SigLIP2-B/16, Precision=W8A8, Evaluation Protocol=Zero-shot2025.1069.74————
RepQ-ViTBackbone=SigLIP2-B/16, Precision=W8A8, Evaluation Protocol=Zero-shot2025.1069.2————
MobileCLIP-BEvaluation Protocol=zero-shot2023.1168.888.392.9——
FIMA-QBackbone=SigLIP2-B/16, Precision=W4A4, Evaluation Protocol=Zero-shot2025.1067.7————
FIMA-QBackbone=SigLIP2-B/16, Precision=W8A8, Evaluation Protocol=Zero-shot2025.1067.7————
FIMA-QBackbone=SigLIP2-B/16, Precision=W6A6, Evaluation Protocol=Zero-shot2025.1067.68————
ViCLIP-B/16Evaluation Protocol=zero-shot2023.1167.287.392.7——
ERQ w/ RegCacheBackbone=SigLIP2-B/16, Precision=W6A6, Evaluation Protocol=Zero-shot2025.1065.56————
CLIP-ERMoE2025.1165.488.394.7——
CLIP-MoE2025.11658692——
ERQBackbone=SigLIP2-B/16, Precision=W6A6, Evaluation Protocol=Zero-shot2025.1064.8————
Naïve w/ RegCacheBackbone=SigLIP2-B/16, Precision=W8A8, Evaluation Protocol=Zero-shot2025.1064.02————
MobileCLIP-S2Evaluation Protocol=zero-shot2023.1163.485.191.4——
Long-CLIP2025.1162.885.191.2——
Fix-CLIP2025.1162.385.491.4——
MobileCLIP-S1Evaluation Protocol=zero-shot2023.1162.284.390.1——
NoisyQuantBackbone=SigLIP2-B/16, Precision=W8A8, Evaluation Protocol=Zero-shot2025.1062.04————
FP32Backbone=OpenCLIP-B/16, Precision=FP32, Evaluation Protocol=Zero-shot2025.1061.02————
PTQ4ViT w/ RegCacheBackbone=OpenCLIP-B/16, Precision=W6A6, Evaluation Protocol=Zero-shot2025.1060————
PTQ4ViT w/ RegCacheBackbone=OpenCLIP-B/16, Precision=W8A8, Evaluation Protocol=Zero-shot2025.1060————
FIMA-Q w/ RegCacheBackbone=OpenCLIP-B/16, Precision=W6A6, Evaluation Protocol=Zero-shot2025.1059.88————
FIMA-Q w/ RegCacheBackbone=OpenCLIP-B/16, Precision=W8A8, Evaluation Protocol=Zero-shot2025.1059.86————
PTQ4ViTBackbone=OpenCLIP-B/16, Precision=W8A8, Evaluation Protocol=Zero-shot2025.1059.6————
ERQ w/ RegCacheBackbone=OpenCLIP-B/16, Precision=W8A8, Evaluation Protocol=Zero-shot2025.1059.56————
RepQ-ViT w/ RegCacheBackbone=OpenCLIP-B/16, Precision=W8A8, Evaluation Protocol=Zero-shot2025.1059.44————
FIMA-QBackbone=OpenCLIP-B/16, Precision=W8A8, Evaluation Protocol=Zero-shot2025.1059.26————
B-cosifiedMode=Zero-shot2025.1259.181.1———
OursMode=Zero-shot2025.1258.980.9———
FIMA-QBackbone=OpenCLIP-B/16, Precision=W6A6, Evaluation Protocol=Zero-shot2025.1058.84————
RepQ-ViT w/ RegCacheBackbone=SigLIP2-B/16, Precision=W6A6, Evaluation Protocol=Zero-shot2025.1058.78————
ERQBackbone=OpenCLIP-B/16, Precision=W8A8, Evaluation Protocol=Zero-shot2025.1058.74————
MobileCLIP-S0Evaluation Protocol=zero-shot2023.1158.781.188.2——
CLIPMode=Zero-shot2025.1258.780.7———
RepQ-ViTBackbone=OpenCLIP-B/16, Precision=W8A8, Evaluation Protocol=Zero-shot2025.1057.62————
Naïve w/ RegCacheBackbone=OpenCLIP-B/16, Precision=W8A8, Evaluation Protocol=Zero-shot2025.1057.6————
FIMA-Q w/ RegCacheBackbone=OpenCLIP-B/16, Precision=W4A4, Evaluation Protocol=Zero-shot2025.1057.34————
RepQ-ViTBackbone=SigLIP2-B/16, Precision=W6A6, Evaluation Protocol=Zero-shot2025.1057————
TinyCLIP-63M/32Evaluation Protocol=zero-shot2023.1156.9————
NoisyQuant w/ RegCacheBackbone=OpenCLIP-B/16, Precision=W8A8, Evaluation Protocol=Zero-shot2025.1056.06————
FIMA-QBackbone=OpenCLIP-B/16, Precision=W4A4, Evaluation Protocol=Zero-shot2025.1055.82————
PTQ4ViT w/ RegCacheBackbone=SigLIP2-B/16, Precision=W6A6, Evaluation Protocol=Zero-shot2025.1053.9————
NoisyQuantBackbone=OpenCLIP-B/16, Precision=W8A8, Evaluation Protocol=Zero-shot2025.1053.84————
CLIPA-B/16Evaluation Protocol=zero-shot2023.1153.1————
NoisyQuant w/ RegCacheBackbone=OpenCLIP-B/16, Precision=W6A6, Evaluation Protocol=Zero-shot2025.1051.58————
NoisyQuant w/ RegCacheBackbone=SigLIP2-B/16, Precision=W6A6, Evaluation Protocol=Zero-shot2025.1050.66————
PTQ4ViTBackbone=OpenCLIP-B/16, Precision=W6A6, Evaluation Protocol=Zero-shot2025.1050.6————
NoisyQuantBackbone=OpenCLIP-B/16, Precision=W6A6, Evaluation Protocol=Zero-shot2025.1043.86————
NaïveBackbone=OpenCLIP-B/16, Precision=W8A8, Evaluation Protocol=Zero-shot2025.1037.32————
ERQ w/ RegCacheBackbone=OpenCLIP-B/16, Precision=W6A6, Evaluation Protocol=Zero-shot2025.1033.64————
RepQ-ViT w/ RegCacheBackbone=OpenCLIP-B/16, Precision=W6A6, Evaluation Protocol=Zero-shot2025.1032.96————
ERQBackbone=OpenCLIP-B/16, Precision=W6A6, Evaluation Protocol=Zero-shot2025.1032.56————
PTQ4ViTBackbone=SigLIP2-B/16, Precision=W6A6, Evaluation Protocol=Zero-shot2025.1029.88————
NoisyQuantBackbone=SigLIP2-B/16, Precision=W6A6, Evaluation Protocol=Zero-shot2025.1028.74————
SGRNoise=20%2024.0525.758.875.1317.1—
FLIPAttnmode=Zero-shot2024.0523.8948.3360.04——
Ours-Embeddingstrategy=Embedding, mode=Zero-shot2024.0523.7748.1859.76——
Ours-RGB0.3strategy=RGB, anchor_patch_ratio=0.3, mode=Zero-shot2024.0523.6547.5459.25——
CLIPmode=Zero-shot2024.0523.4947.859.66——
Ours-RGB0.5strategy=RGB, anchor_patch_ratio=0.5, mode=Zero-shot2024.0522.9747.2959.21——
Ours-Kmeansstrategy=Kmeans, mode=Zero-shot2024.0522.646.9358.84——