List your company
ResearchFormerly WizWand

Image Classification on SVHN (Top-1 Accuracy)

98.21Top-1 AccuracyθB fine-tune
92.198893.759495.3296.8806Mar 29, 2025Jun 8, 2025Aug 19, 2025Oct 30, 2025Jan 10, 2026Mar 23, 2026Jun 3, 2026
Updated 5mo ago

Evaluation results

MethodLinks
θB fine-tuneTarget Backbone=ViT-L/14, Target Pre-training=Datacomp-XL, Evaluation Protocol=fine-tune2026.0598.21
θA fine-tuneNumber of calibration samples (N)=-2026.0598.21
CPolyNeXt-LRtraining=from scratch, resolution=native resolution, activation-free=true2026.0598.1
θB fine-tuneNumber of calibration samples (N)=-2026.0598.02
θA fine-tuneSource Backbone=ViT-B/16, Source Pre-training=LAION-2B, Evaluation Protocol=fine-tune2026.0597.89
RDCNet2026.0497.78
DAMSNet2026.0497.6
MONet-Ttraining=from scratch, resolution=native resolution, activation-free=true2026.0597.6
IndividualBackbone=ViT-B/322026.0497.46
Auto-FlexSwitch+CKA (λ = 3)Type=Dynamic, Storage (MB)=47.26, Backbone=ViT-B/322026.0497.45
SSCNet2026.0497.43
Twin-MergingType=Dynamic, Storage (MB)=711.10, Backbone=ViT-B/322026.0497.36
Adam-DualOpt_Fine-tuneBackbone=ViT-B2026.0497.35
Auto-FlexSwitch+MSE (λ = 0.05)Type=Dynamic, Storage (MB)=50.72, Backbone=ViT-B/322026.0497.33
Auto-SwitchType=Dynamic, Storage (MB)=185.28, Backbone=ViT-B/322026.0497.31
Fine-tuned (STL)Backbone=CLIP-ViT-B/32, Bit-width=Full-precision, Training=Single-task learning2026.0597.3
WUDI-Merging w/ E-PMQMerging Method=WUDI-Merging, Quantization Method=E-PMQ, Bit-width=4-bit, Model Backbone=CLIP-ViT-L/142026.0597.3
ResNet-18training=from scratch, resolution=native resolution, activation-free=false2026.0597.3
FullBackbone=ViT-B2026.0497.27
MoW-MergingType=Dynamic, Storage (MB)=2171.89, Backbone=ViT-B/322026.0497.21
QKFormer2026.0497.13
L2-SPBackbone=ViT-B2026.0497.12
SGD-DualOpt_Fine-tuneBackbone=ConvNeXt-B2026.0497.12
Auto-FlexSwitch+KL (λ = 0.3)Type=Dynamic, Storage (MB)=41.51, Backbone=ViT-B/322026.0497.11
Simple Averaging w/ E-PMQMerging Method=Simple Averaging, Quantization Method=E-PMQ, Bit-width=4-bit, Model Backbone=CLIP-ViT-L/142026.0597.1
ResMLPtraining=from scratch, resolution=native resolution, activation-free=false2026.0597.1
FFT MoE# Params (%)=770, Backbone=CLIP ViT-B/322026.0497.09
L2-SPBackbone=ConvNeXt-B2026.0497.01
Task Arithmetic w/ E-PMQMerging Method=Task Arithmetic, Quantization Method=E-PMQ, Bit-width=4-bit, Model Backbone=CLIP-ViT-L/142026.0597
TIES-Merging w/ E-PMQMerging Method=TIES-Merging, Quantization Method=E-PMQ, Bit-width=4-bit, Model Backbone=CLIP-ViT-L/142026.0597
WUDI-MergingMerging Method=WUDI-Merging, Quantization Method=None, Bit-width=Full-precision, Model Backbone=CLIP-ViT-L/142026.0597
FullBackbone=ConvNeXt-B2026.0496.97
WUDI-Merging w/ GPTQMerging Method=WUDI-Merging, Quantization Method=GPTQ, Bit-width=4-bit, Model Backbone=CLIP-ViT-L/142026.0596.9
LoRABackbone=ConvNeXt-B2026.0496.86
EMR-MergingType=Dynamic, Storage (MB)=540.19, Backbone=ViT-B/322026.0496.86
MLP-Mixertraining=from scratch, resolution=native resolution, activation-free=false2026.0596.8
FFT# Params (%)=100, Backbone=CLIP ViT-B/322026.0496.78
FourierMoE# Params (%)=0.42, Backbone=CLIP ViT-B/322026.0496.78
LoRABackbone=ViT-B2026.0496.69
Traditional MTLBackbone=ViT-B/322026.0496.67
GOAT# Params (%)=2.24, Backbone=CLIP ViT-B/322026.0496.62
WUDI-Merging w/ RTNMerging Method=WUDI-Merging, Quantization Method=RTN, Bit-width=4-bit, Model Backbone=CLIP-ViT-L/142026.0596.6
WRN-28-102026.0496.58
LoRA (rank32)# Params (%)=5.98, Backbone=CLIP ViT-B/322026.0496.55
ZAQBackbone=MobileNet-V2, Bit-width=4-bit2026.0696.3
EmaQBackbone=MobileNet-V2, Bit-width=4-bit2026.0696.3
Choi et al.Backbone=MobileNet-V2, Bit-width=4-bit2026.0696.2
TeacherBackbone=AlexNet2026.0496.16
USMoESize=110M2025.0396.1
HydraLoRA# Params (%)=1.58, Backbone=CLIP ViT-B/322026.0496.06
LoRA (rank16)# Params (%)=2.99, Backbone=CLIP ViT-B/322026.0496
MoLoRA# Params (%)=2.24, Backbone=CLIP ViT-B/322026.0496
DIP-KDBackbone=AlexNet, Synthetic Dataset Size=50K2026.0495.94
PiSSA# Params (%)=1.49, Backbone=CLIP ViT-B/322026.0495.84
AdaMoLE# Params (%)=2.33, Backbone=CLIP ViT-B/322026.0495.82
ZeroQBackbone=MobileNet-V2, Bit-width=4-bit2026.0695.8
EmaQBackbone=MobileNet-V2, Bit-width=2-bit2026.0695.7
HO-ResNet2026.0495.69
USMoESize=10M2025.0395.6
ZeroQBackbone=ResNet-20, Bit-width=4-bit2026.0695.6
Choi et al.Backbone=ResNet-20, Bit-width=4-bit2026.0695.6
EmaQBackbone=ResNet-20, Bit-width=4-bit2026.0695.6
ResNet-342026.0495.51
MiLoRA# Params (%)=1.49, Backbone=CLIP ViT-B/322026.0495.49
DoRA# Params (%)=1.49, Backbone=CLIP ViT-B/322026.0495.46
DFHL-RSBackbone=AlexNet, Synthetic Dataset Size=50K2026.0495.41
LoRA (rank8)# Params (%)=1.49, Backbone=CLIP ViT-B/322026.0495.39
DF-MergeType=Static, Backbone=ViT-B/322026.0495.37
WUDI-Merging w/ E-PMQBackbone=CLIP-ViT-B/32, Merging Strategy=WUDI-Merging, Quantization Algorithm=E-PMQ, Bit-width=4-bit2026.0595.3
ATONet2026.0495.21
ZAQBackbone=ResNet-20, Bit-width=4-bit2026.0695.2
EmaQBackbone=ResNet-20, Bit-width=2-bit2026.0695.2
TCSize=10M2025.0395.1
ECSize=10M2025.0395
CAPR-DenseNet2026.0494.97
Simple Averaging w/ E-PMQBackbone=CLIP-ViT-B/32, Merging Strategy=Simple Averaging, Quantization Algorithm=E-PMQ, Bit-width=4-bit2026.0594.9
WUDI-Merging w/ AWQMerging Method=WUDI-Merging, Quantization Method=AWQ, Bit-width=4-bit, Model Backbone=CLIP-ViT-L/142026.0594.9
ZAQBackbone=ResNet-20, Bit-width=2-bit2026.0694.9
Task Arithmetic w/ E-PMQBackbone=CLIP-ViT-B/32, Merging Strategy=Task Arithmetic, Quantization Algorithm=E-PMQ, Bit-width=4-bit, Lambda=0.32026.0594.8
WUDI-MergingBackbone=CLIP-ViT-B/32, Merging Strategy=WUDI-Merging, Bit-width=Full-precision2026.0594.8
WUDI-Merging w/ GPTQBackbone=CLIP-ViT-B/32, Merging Strategy=WUDI-Merging, Quantization Algorithm=GPTQ, Bit-width=4-bit2026.0594.8
SoftMoESize=10M2025.0394.8
TIES-Merging w/ E-PMQBackbone=CLIP-ViT-B/32, Merging Strategy=TIES-Merging, Quantization Algorithm=E-PMQ, Bit-width=4-bit, Lambda=0.32026.0594.7
Choi et al.Backbone=ResNet-20, Bit-width=2-bit2026.0694.7
Multi-ResNet2026.0494.58
TCSize=110M2025.0394.5
VPTBackbone=ViT-B2026.0494.37
ZeroQBackbone=ResNet-20, Bit-width=2-bit2026.0694.3
MMA-CCT2026.0494.29
Couplformer2026.0494.26
WUDI-Merging w/ RTNBackbone=CLIP-ViT-B/32, Merging Strategy=WUDI-Merging, Quantization Algorithm=RTN, Bit-width=4-bit2026.0594.2
WUDI-Merging w/ AWQBackbone=CLIP-ViT-B/32, Merging Strategy=WUDI-Merging, Quantization Algorithm=AWQ, Bit-width=4-bit2026.0593.9
GhostNet2026.0493.86
SoftMoESize=110M2025.0393.5
LLSQBackbone=ResNet-20, Bit-width=4-bit2026.0693.4
EfficientNet2026.0493.34
FAVOR+2026.0493.21
ECSize=110M2025.0393.2
LLSQBackbone=ResNet-20, Bit-width=2-bit2026.0693
WUDI-Merging w/ E-PMQMerging Strategy=WUDI-Merging, Quantization Method=E-PMQ, Model Backbone=CLIP-ViT-B/32, Bit-width=4-bit2026.0592.43