List your company
ResearchFormerly WizWand

Fine-grained Image Classification on DTD (test)

15.6Top-1 ErrorFull FT
15.33617.11818.920.682Mar 7, 2023
Updated 4mo ago

Evaluation results

MethodLinks
Full FTBackbone=SWIN-L, % Trainable parameters=100%, No backbone backpropagation=false2023.0315.6——
LoRABackbone=SWIN-L, % Trainable parameters=0.8%, No backbone backpropagation=false2023.0315.8——
In. MLP-3Backbone=SWIN-L, % Trainable parameters=2.8%, No backbone backpropagation=true2023.0316.7——
MLP-3Backbone=SWIN-L, % Trainable parameters=2.8%, No backbone backpropagation=false2023.0316.7——
AdaLNBackbone=SWIN-L, % Trainable parameters=0.1%, No backbone backpropagation=false2023.0316.7——
BitFitBackbone=SWIN-L, % Trainable parameters=0.1%, No backbone backpropagation=false2023.0317——
InCABackbone=ViT-L/162023.0317.2——
In. MLP-3Backbone=ViT-L/162023.0317.4——
In. LPBackbone=SWIN-L, % Trainable parameters=0.1%, No backbone backpropagation=true2023.0317.7——
InCABackbone=SWIN-L, % Trainable parameters=3.7%, No backbone backpropagation=true2023.0317.8——
Full FTBackbone=ViT-L/162023.0318.2——
InCA (last)Backbone=ViT-L/162023.0318.4——
In. LPBackbone=ViT-L/162023.0318.9——
InCA (last)Backbone=SWIN-L, % Trainable parameters=3.7%, No backbone backpropagation=true2023.0319.1——
LPBackbone=SWIN-L, % Trainable parameters=0.1%, No backbone backpropagation=false2023.0319.1——
LoRABackbone=ViT-L/162023.0319.4——
MLP-3Backbone=ViT-L/162023.0320.1——
LPBackbone=ViT-L/162023.0320.6——
VPTBackbone=ViT-L/162023.0321.4——
BitFitBackbone=ViT-L/162023.0321.9——
AdaLNBackbone=ViT-L/162023.0322.2——
CLIPCategory=Baseline, Backbone=CLIP-ViT-B/32, Attack Protocol=PGD-100 (ϵ = 4.0)2026.06—42.10
CLIPBackbone=CLIP-ResNet50, Epsilon=1.02026.06—40.42.1
DOCBackbone=CLIP-ResNet50, Epsilon=1.0, Effective batch size=12026.06—38.515.5
EnsembleBackbone=CLIP-ResNet50, Epsilon=1.0, Augmentation views=152026.06—37.825.8
MACCategory=Tuning-free, Backbone=CLIP-ViT-B/32, Attack Protocol=PGD-100 (ϵ = 4.0)2026.06—41.732.9
MTACategory=Tuning-free, Backbone=CLIP-ViT-B/32, Attack Protocol=PGD-100 (ϵ = 4.0)2026.06—43.215.7
R-TPTCategory=Tuning-based, Backbone=CLIP-ViT-B/32, Attack Protocol=PGD-100 (ϵ = 4.0)2026.06—42.129.3
R-TPTBackbone=CLIP-ResNet50, Epsilon=1.0, Augmentation views=152026.06—39.127.1
SS-TPTBackbone=CLIP-ResNet50, Epsilon=1.0, Augmentation views=152026.06—4034.6
TAPTCategory=Tuning-based, Backbone=CLIP-ViT-B/32, Attack Protocol=PGD-100 (ϵ = 4.0)2026.06—39.212.4
TPTBackbone=CLIP-ResNet50, Epsilon=1.0, Augmentation views=152026.06—41.34.5
TTCCategory=Tuning-free, Backbone=CLIP-ViT-B/32, Attack Protocol=PGD-100 (ϵ = 4.0)2026.06—38.96.2
TTCBackbone=CLIP-ResNet50, Epsilon=1.0, Effective batch size=12026.06—38.615.2