List your company
ResearchFormerly WizWand

Image Classification on EUROSAT (Accuracy %)

99Accuracytheta_B fine-tune
9.40432.664555.92579.1855Oct 7, 2025Nov 13, 2025Dec 21, 2025Jan 28, 2026Mar 6, 2026Apr 13, 2026May 21, 2026
Updated 5mo ago

Evaluation results

MethodLinks
theta_B fine-tuneModel=ViT-L/14, Mode=fine-tune2026.0299
theta_B fine-tuneModel=laion400m, Protocol=fine-tune2026.0298.96
theta_B fine-tuneBackbone=ViT-L/142025.1098.95
theta_B fine-tuneBackbone=ViT-B/162025.1098.7
theta_A fine-tuneModel=laion2b, Protocol=fine-tune2026.0298.69
theta_A fine-tuneModel=ViT-B/16, Mode=fine-tune2026.0298.66
theta_B + delta_starBackbone=ViT-L/142025.1096.75
theta_B + delta_starBackbone=ViT-B/162025.1095.06
THESEUSModel=ViT-L/14, B (alignment batches)=102026.0276.78
GradFix (theta_B + delta_A)Backbone=ViT-L/14, |Ds_c|=52025.1075.59
GradFix (theta_B + delta_A)Backbone=ViT-L/14, |Ds_c|=22025.1074.1
theta_B_opt (Few-shot fine-tuning)Backbone=ViT-L/14, |Ds_c|=22025.1070.76
theta_B_opt (Few-shot fine-tuning)Backbone=ViT-L/14, |Ds_c|=52025.1069.75
GradFix (theta_B + delta_A)Backbone=ViT-L/14, |Ds_c|=12025.1069.67
theta_B zero-shotModel=ViT-L/14, Mode=zero-shot2026.0268.4
THESEUSK=20, Width Scaling Direction=laion2b to laion400m2026.0266.83
GradFix (theta_B + delta_A)Backbone=ViT-B/16, |Ds_c|=52025.1066.05
THESEUSK=10, Width Scaling Direction=laion2b to laion400m2026.0265.51
GradFix (theta_B + delta_A)Backbone=ViT-B/16, |Ds_c|=22025.1065.07
theta_B_opt (Few-shot fine-tuning)Backbone=ViT-L/14, |Ds_c|=12025.1064.65
THESEUSK=5, Width Scaling Direction=laion2b to laion400m2026.0264.58
TransFusionBackbone=ViT-L/142025.1063.21
theta_B zero-shotBackbone=ViT-L/142025.1062.8
theta_B + tau_ABackbone=ViT-L/142025.1062.77
theta_B_opt (Few-shot fine-tuning)Backbone=ViT-B/16, |Ds_c|=52025.1061.99
GradFix (theta_B + delta_A)Backbone=ViT-B/16, |Ds_c|=12025.1061.94
THESEUSK=2, Width Scaling Direction=laion2b to laion400m2026.0261.46
THESEUSK=1, Width Scaling Direction=laion2b to laion400m2026.0260.79
theta_B_opt (Few-shot fine-tuning)Backbone=ViT-B/16, |Ds_c|=22025.1059.49
theta_B_opt (Few-shot fine-tuning)Backbone=ViT-B/16, |Ds_c|=12025.1056.61
CLIP ViT-B/16Img. encoder=CLIP ViT-B/162026.0551.93
theta_B zero-shotModel=laion400m, Protocol=zero-shot2026.0250.92
TransFusionBackbone=ViT-B/162025.1050.12
theta_B + tau_ABackbone=ViT-B/162025.1049.58
theta_B zero-shotBackbone=ViT-B/162025.1049.41
CLIP ViT-B/32Img. encoder=CLIP ViT-B/322026.0534.44
CLIP RN-50Img. encoder=CLIP RN-502026.0528
ConvFormer-S18 + MPNetImg. encoder=ConvFormer-S18, Txt. encoder=MPNet2026.0527
CLIP RN-101Img. encoder=CLIP RN-1012026.0526.7
TinyViT-21M + CLIPImg. encoder=TinyViT-21M, Txt. encoder=CLIP2026.0526.48
TinyViT-21M + MPNetImg. encoder=TinyViT-21M, Txt. encoder=MPNet2026.0525.56
TinyViT-21M + MiniLMImg. encoder=TinyViT-21M, Txt. encoder=MiniLM2026.0524.93
ConvNext-B + CLIPImg. encoder=ConvNext-B, Txt. encoder=CLIP2026.0524.78
ConvFormer-S18 + CLIPImg. encoder=ConvFormer-S18, Txt. encoder=CLIP2026.0523.93
BEiT-B/16 + CLIPImg. encoder=BEiT-B/16, Txt. encoder=CLIP2026.0523.67
ConvNext-B + MPNetImg. encoder=ConvNext-B, Txt. encoder=MPNet2026.0523.63
CAFormer-S18 + CLIPImg. encoder=CAFormer-S18, Txt. encoder=CLIP2026.0522.7
TinyViT-21M + RoBERTaImg. encoder=TinyViT-21M, Txt. encoder=RoBERTa2026.0521.48
CAFormer-S18 + RoBERTaImg. encoder=CAFormer-S18, Txt. encoder=RoBERTa2026.0520.7
CAFormer-S18 + MPNetImg. encoder=CAFormer-S18, Txt. encoder=MPNet2026.0520.48
ConvNext-B + RoBERTaImg. encoder=ConvNext-B, Txt. encoder=RoBERTa2026.0519.15
ConvFormer-S18 + MiniLMImg. encoder=ConvFormer-S18, Txt. encoder=MiniLM2026.0518.59
BEiT-B/16 + MPNetImg. encoder=BEiT-B/16, Txt. encoder=MPNet2026.0518.15
CAFormer-S18 + MiniLMImg. encoder=CAFormer-S18, Txt. encoder=MiniLM2026.0517.37
BEiT-B/16 + RoBERTaImg. encoder=BEiT-B/16, Txt. encoder=RoBERTa2026.0517.15
ConvNext-B + MiniLMImg. encoder=ConvNext-B, Txt. encoder=MiniLM2026.0517
ConvFormer-S18 + RoBERTaImg. encoder=ConvFormer-S18, Txt. encoder=RoBERTa2026.0516.44
BEiT-B/16 + MiniLMImg. encoder=BEiT-B/16, Txt. encoder=MiniLM2026.0512.85