List your company
ResearchFormerly WizWand

Classification on Caltech101

97.6AccuracyFLYP
23.44842.69961.9581.201Oct 17, 2022May 31, 2023Jan 12, 2024Aug 25, 2024Apr 8, 2025Nov 20, 2025Jul 4, 2026
Updated 3mo ago

Evaluation results

MethodLinks
FLYPEvaluation Protocol=Fine-tuning2022.1297.6
FTEvaluation Protocol=Fine-tuning2022.1297.2
LP-FTEvaluation Protocol=Linear Probing then Fine-tuning2022.1296.9
ProposedShots=16, Venue=-2025.1296.2
TPTBackbone=ViT-B/162026.0694.96
LPEvaluation Protocol=Linear Probing2022.1294.8
USEBackbone=ResNet50, Optim.=true2026.0787.95
RLCFBackbone=ResNet50, Optim.=true2026.0787.83
TPTBackbone=ResNet50, Optim.=true2026.0787.77
SEBackbone=ResNet50, Optim.=false2026.0787.75
ZeroshotEvaluation Protocol=Zero-shot2022.1287.7
MTABackbone=ResNet50, Optim.=false2026.0787.46
C-TPTBackbone=ResNet50, Optim.=true2026.0787.34
TPTBackbone=ResNet-502026.0687.02
TPSBackbone=ResNet50, Optim.=true2026.0786.96
STSBackbone=ResNet50, Optim.=true2026.0786.63
CLIPShots=0, Venue=ICML’222025.1286.5
ZEROBackbone=ResNet50, Optim.=false2026.0786.43
R-TPTBackbone=ResNet50, Optim.=true2026.0786.13
CLIPBackbone=ResNet50, Optim.=false2026.0785.88
LSA+OnZetaBackbone=ViT-B/162026.0684.72
LSABackbone=ViT-B/162026.0684.46
CLIP baselineBackbone=ViT-B/162026.0683.91
OnZetaBackbone=ViT-B/162026.0683.86
Noun SubmanifoldBackbone=CLIP ViT-L-14, zero-shot=true2023.0582.8
CLIPBackbone=CLIP, Expansion Ratio=1x, Evaluation Protocol=Zero-shot2022.1182.1
CLIPBackbone=CLIP ViT-L-14, zero-shot=true2023.0580.9
POS PCABackbone=CLIP ViT-L-14, zero-shot=true2023.0580.9
POS PGABackbone=CLIP ViT-L-14, zero-shot=true2023.0580.8
LSA+OnZetaBackbone=ResNet-502026.0679.63
OnZetaBackbone=ResNet-502026.0679.39
CLIP baselineBackbone=ResNet-502026.0679.14
LSABackbone=ResNet-502026.0678.82
xCLIPBackbone=ViT-B/16, Pre-trained on=IT35M, Mode=Zero-shot2022.1073.9
E-CLIPImage Encoder=ViT-B/16, Protocol=Zero-shot2025.0373.7
CLIPImage Encoder=ViT-B/16, Protocol=Zero-shot2025.0373.5
A-CLIPImage Encoder=ViT-B/16, Protocol=Zero-shot2025.0373.5
CLIP-PGSImage Encoder=ViT-B/16, p=0.3, Protocol=Zero-shot2025.0372.7
CLIP-PGSImage Encoder=ViT-B/16, p=0.5, Protocol=Zero-shot2025.0371.8
CLIP-PGSImage Encoder=ViT-S/16, p=0.3, Protocol=Zero-shot2025.0371.8
CLIPBackbone=ViT-B/16, Pre-trained on=IT35M, Mode=Zero-shot2022.1071.3
FLIPImage Encoder=ViT-B/16, Protocol=Zero-shot2025.0370.4
CLIP-PGSImage Encoder=ViT-S/16, p=0.5, Protocol=Zero-shot2025.0369.4
GIF-SDBackbone=ResNet-50, Expansion Method=GIF-SD, Expansion Ratio=20x, Evaluation Protocol=Train from scratch2022.1165.1
nCLIPBackbone=ViT-B/16, Pre-trained on=IT35M, Mode=Zero-shot2022.1065
GIF-DALLEBackbone=ResNet-50, Expansion Method=GIF-DALLE, Expansion Ratio=20x, Evaluation Protocol=Train from scratch2022.1163
DALL-E2Backbone=ResNet-50, Expansion Method=DALL-E2, Expansion Ratio=20x, Evaluation Protocol=Train from scratch2022.1161.3
FLQMIAAlignment=frozen pretrained DINOv2 and GTE-1.5, Setting=Zero-shot2026.0559.06
GIF-MAEBackbone=ResNet-50, Expansion Method=GIF-MAE, Expansion Ratio=20x, Evaluation Protocol=Train from scratch2022.1158.4
SAIL w/ SigLIPAlignment=frozen pretrained DINOv2 and GTE-1.5, Loss=SigLIP, Setting=Zero-shot2026.0558.23
RandAugmentBackbone=ResNet-50, Expansion Method=RandAugment, Expansion Ratio=20x, Evaluation Protocol=Train from scratch2022.1157.8
SAIL w/ CLIPAlignment=frozen pretrained DINOv2 and GTE-1.5, Loss=InfoNCE, Setting=Zero-shot2026.0555.66
FLVMIAAlignment=frozen pretrained DINOv2 and GTE-1.5, Setting=Zero-shot2026.0553.3
GridMaskBackbone=ResNet-50, Expansion Method=GridMask, Expansion Ratio=20x, Evaluation Protocol=Train from scratch2022.1151.6
CutoutBackbone=ResNet-50, Expansion Method=Cutout, Expansion Ratio=20x, Evaluation Protocol=Train from scratch2022.1151.5
SDBackbone=ResNet-50, Expansion Method=SD, Expansion Ratio=20x, Evaluation Protocol=Train from scratch2022.1151.1
MAEBackbone=ResNet-50, Expansion Method=MAE, Expansion Ratio=20x, Evaluation Protocol=Train from scratch2022.1150.6
CSAAlignment=frozen pretrained DINOv2 and GTE-1.5, Method=Analytical (CCA), Setting=Zero-shot2026.0546.28
Distillation of CLIPBackbone=ResNet-50, Expansion Ratio=1x, Evaluation Protocol=Knowledge Distillation2022.1133.2
OriginalBackbone=ResNet-50, Expansion Ratio=1x, Evaluation Protocol=Train from scratch2022.1126.3