List your company
ResearchFormerly WizWand

Image Classification on Oxford Flowers-102 (test)

99.847Top-1 AccuracyEfficient Adaptive Ensembling
89.7517292.3726194.993597.61439Nov 16, 2018Feb 8, 2020May 2, 2021Jul 26, 2022Oct 18, 2023Jan 9, 2025Apr 4, 2026
Updated 4mo ago

Evaluation results

MethodLinks
Efficient Adaptive Ensemblingensemble size=2 weak models2022.0699.8470.127—
ViT-L/16 + BackgroundArchitecture=ViT-L/16, Epochs=2, Spinal FC=false, Background Class=true2023.0599.75——
Vision Transformer (ViT-L/16)Pre-training Dataset=JFT-300M, Architecture=ViT-L/162020.1099.74——
SOTA2022.0699.72——
CvT-W24Param (M)=277, Pre-trained=ImageNet-22k2021.0399.72——
Vision Transformer (ViT-H/14)Pre-training Dataset=JFT-300M, Architecture=ViT-H/142020.1099.68——
BiT-LArchitecture=ResNet152x42020.1099.63——
CvT-21Param (M)=32, Pre-trained=ImageNet-22k2021.0399.62——
ViT-L/16Param (M)=307, Pre-trained=ImageNet-22k2021.0399.61——
Vision Transformer (ViT-L/16)Pre-training Dataset=ImageNet-21k, Architecture=ViT-L/162020.1099.61——
ViT-L/16 + (Spinal FC & Background)Architecture=ViT-L/16, Epochs=2, Spinal FC=true, Background Class=true2023.0599.6——
ViT-H/16Param (M)=632, Pre-trained=ImageNet-22k2021.0399.51——
ViT-L/16Architecture=ViT-L/16, Epochs=2, Spinal FC=false, Background Class=false2023.0599.51——
CvT-13Param (M)=20, Pre-trained=ImageNet-22k2021.0399.5——
ViT-L/16 + Spinal FCArchitecture=ViT-L/16, Epochs=2, Spinal FC=true, Background Class=false2023.0599.41——
ViT-B/16Param (M)=86, Pre-trained=ImageNet-22k2021.0399.38——
BiT-MParam (M)=928, Pre-trained=ImageNet-22k2021.0399.3——
Grafit RegNetY-8GFParams (M)=39.2, Resolution=384x384, Pre-training=ImageNet2021.0299.1——
Graph-PruneEpochs=100, Backbone=EfficientNet-B0, Pruning strategy=Eigenvector Centrality, Layer pruned=Final Dense layer2025.1299.07——
TNT-BParams (M)=65.6, Resolution=384x384, Pre-training=ImageNet2021.0299——
iBOTBackbone=ViT-B/16, Evaluation Protocol=Fine-tuning2021.1198.9——
DeiT-BGPU Throughput (images/sec)=114, Resolution=384x384, Fine-tuned=true2026.0398.9——
LowFormer-B3GPU Throughput (images/sec)=424, Resolution=384x384, Fine-tuned=true2026.0398.9——
GPipeParams=556M, Pre-trained=ImageNet ILSVRC2012, Evaluation Protocol=finetuned2021.0498.8——
EfficientNet-B7Params=66M, Pre-trained=ImageNet ILSVRC2012, Evaluation Protocol=finetuned2021.0498.8——
EfficientNetV2-LParams=121M, Pre-trained=ImageNet ILSVRC2012, Evaluation Protocol=finetuned2021.0498.8——
TNT-SParams (M)=23.8, Resolution=384x384, Pre-training=ImageNet2021.0298.8——
DINOBackbone=ViT-B/16, Evaluation Protocol=Fine-tuning2021.1198.8——
DINOprotocol=Fine Tuning, arch.=ViT-B/162024.0898.8——
MoCo v3Backbone=ViT-H, Pre-training=ImageNet-1k Self-Supervised, Evaluation Protocol=End-to-end fine-tuning2021.0498.8——
TNT-SGPU Throughput (images/sec)=141, Resolution=384x384, Fine-tuned=true2026.0398.8——
iBOTBackbone=ViT-S/16, Evaluation Protocol=Fine-tuning2021.1198.6——
MoCo v3Backbone=ViT-L, Pre-training=ImageNet-1k Self-Supervised, Evaluation Protocol=End-to-end fine-tuning2021.0498.6——
CeiT-SGPU Throughput (images/sec)=260, Resolution=384x384, Fine-tuned=true2026.0398.6——
DeiT-B-384Params=86M, Pre-trained=ImageNet ILSVRC2012, Evaluation Protocol=finetuned2021.0498.5——
EfficientNetV2-MParams=55M, Pre-trained=ImageNet ILSVRC2012, Evaluation Protocol=finetuned2021.0498.5——
EfficientNet-B5Params (M)=30, Resolution=384x384, Pre-training=ImageNet2021.0298.5——
EfficientNet-B5Params (M)=30, Fine-tuning=true2021.0698.5——
DINOBackbone=ViT-S/16, Evaluation Protocol=Fine-tuning2021.1198.5——
EfficientNetV2-MGPU Throughput (images/sec)=277, Resolution=384x384, Fine-tuned=true2026.0398.5——
DeiT-BParams=86M, Pre-trained=ImageNet ILSVRC2012, Evaluation Protocol=finetuned2021.0498.4——
DeiT-BParams (M)=86.4, Resolution=384x384, Pre-training=ImageNet2021.0298.4——
DeiT-BParams (M)=86.6, Fine-tuning=true2021.0698.4——
Rand.Backbone=ViT-B/16, Evaluation Protocol=Fine-tuning2021.1198.4——
Grafit ResNet-50Params (M)=25.6, Resolution=384x384, Pre-training=ImageNet2021.0298.2——
Grafit ResNet-50Params (M)=25.6, Fine-tuning=true2021.0698.2——
Rand.Backbone=ViT-S/16, Evaluation Protocol=Fine-tuning2021.1198.2——
DnCBackbone=ResNet-50, Pre-training dataset=JFT, Pre-training epochs=4500 ImageNet-equivalent, Evaluation protocol=Linear transfer2022.0198.2——
DIFFUSEMIXBackbone=ResNet-50, Pre-trained=ImageNet, Evaluation protocol=Fine-tuning2024.0498.02——
BEITBackbone=ViT-B/16, Evaluation Protocol=Fine-tuning2021.1198——
EfficientNetV2-SParams=24M, Pre-trained=ImageNet ILSVRC2012, Evaluation Protocol=finetuned2021.0497.9——
BAMprotocol=Fine Tuning, arch.=ViT-B/162024.0897.9——
ReXNet (x1.0)FLOPs=0.4B, Params=4.8M, Fine-tuning=true2020.0797.8——
ViTAE-SParams (M)=23.6, Fine-tuning=true2021.0697.8——
ImageNet - Adaptive TransferPre-training Method=ImageNet - Adaptive Transfer, Backbone=Inception v32018.1197.7——
EMD SubsetPre-training Method=ImageNet & iNaturalist - EMD Subset, Backbone=Inception v32018.1197.7——
ResNet50FLOPs=4.1B, Params=25.6M, Fine-tuning=true2020.0797.7——
MoCo-v3protocol=Fine Tuning, arch.=ViT-B/162024.0897.7——
MoCo v3Backbone=ViT-B, Pre-training=ImageNet-1k Self-Supervised, Evaluation Protocol=End-to-end fine-tuning2021.0497.7——
ViTAE-TParams (M)=4.8, Fine-tuning=true2021.0697.5——
EfficientNet-B0FLOPs=0.4B, Params=5.3M, Fine-tuning=true2020.0797.3——
AdaAugBackbone=ResNet-50, Pre-trained=ImageNet, Evaluation protocol=Fine-tuning2024.0497.19——
Entire ImageNet DatasetPre-training Method=Entire ImageNet Dataset, Backbone=Inception v32018.1197.1——
Selective Joint FTPre-training Method=ImageNet - Selective Joint FT, Backbone=Inception v32018.1197——
BEITBackbone=ViT-S/16, Evaluation Protocol=Fine-tuning2021.1196.4——
BAMprotocol=Linear Eval., arch.=ResNet-502024.0896.3——
BYOLprotocol=Linear Eval., arch.=ResNet-502024.0896.1——
Fast AABackbone=ResNet-50, Pre-trained=ImageNet, Evaluation protocol=Fine-tuning2024.0496.08——
NETTAILORBackbone=ResNet-50, Input Resolution=224x224, Number of Parameters=8.5M, FLOPs=2.37G2019.0695.79——
ReLICv2Backbone=ResNet-50, Pre-training dataset=JFT, Pre-training epochs=5000 ImageNet-equivalent, Evaluation protocol=Linear transfer2022.0195.7——
RABackbone=ResNet-50, Pre-trained=ImageNet, Evaluation protocol=Fine-tuning2024.0495.23——
NNCLRprotocol=Linear Eval., arch.=ResNet-502024.0895.1——
VanillaBackbone=ResNet-50, Pre-trained=ImageNet, Evaluation protocol=Fine-tuning2024.0494.98——
PiggybackBackbone=ResNet-50, Input Resolution=224x224, Number of Parameters=24.3M, FLOPs=4.11G2019.0694.77——
ReLICv2Backbone=ResNet-50, Pre-training dataset=JFT, Pre-training epochs=1000 ImageNet-equivalent, Evaluation protocol=Linear transfer2022.0194.5——
BYOLBackbone=ResNet-50, Pre-training dataset=JFT, Pre-training epochs=5000 ImageNet-equivalent, Evaluation protocol=Linear transfer2022.0194.3——
AABackbone=ResNet-50, Pre-trained=ImageNet, Evaluation protocol=Fine-tuning2024.0493.88——
REGSLBackbone=ResNet-101, Pre-trained=ImageNet (ILSVRC-2012), Evaluation Protocol=Fine-tuning2021.1193.82——
LSBackbone=ResNet-101, Pre-trained=ImageNet (ILSVRC-2012), Evaluation Protocol=Fine-tuning2021.1193.72——
l2-PGMBackbone=ResNet-101, Pre-trained=ImageNet (ILSVRC-2012), Evaluation Protocol=Fine-tuning2021.1193.23——
l2-SPBackbone=ResNet-101, Pre-trained=ImageNet (ILSVRC-2012), Evaluation Protocol=Fine-tuning2021.1193.06——
PackNet (Forward)Backbone=ResNet-50, Input Resolution=224x224, Order=CUB, Cars, Flowers, WikiArt, Sketch, Number of Parameters=23.9M, FLOPs=4.11G2019.0693.04——
Fine-tuningBackbone=ResNet-101, Pre-trained=ImageNet (ILSVRC-2012), Evaluation Protocol=Fine-tuning2021.1192.92——
l2-NormBackbone=ResNet-101, Pre-trained=ImageNet (ILSVRC-2012), Evaluation Protocol=Fine-tuning2021.1192.76——
ViT-baseParams.=86M, Complexity=O(N^2), Evaluation Protocol=Linear Probing2025.0792.75——
Structured PruningEpochs=100, Backbone=EfficientNet-B0, Pruning strategy=Structured Pruning, Layer pruned=Final Dense layer2025.1292.37——
ResNet50Epochs=100, Backbone=ResNet-50, Pruning strategy=None2025.1291.96——
ViT-B/16-SAMBackbone=ViT-B/16, SAM Pre-training=true2021.0691.8——
ViT-S/16-SAMBackbone=ViT-S/16, SAM Pre-training=true2021.0691.5——
SimCLRprotocol=Linear Eval., arch.=ResNet-502024.0891.2——
ResNet-152-SAMBackbone=ResNet-152, SAM Pre-training=true2021.0691.1——
EfficientNetEpochs=100, Backbone=EfficientNet-B0, Pruning strategy=None2025.1291.04——
EViTKeep rate r=0.7, Backbone=DeiT-S2026.0490.81——
Ours (Rényi, α = 5.0)Keep rate r=0.9, Backbone=DeiT-S2026.0490.71——
Ours (Rényi, α = 2.0)Keep rate r=0.7, Backbone=DeiT-S2026.0490.63——
EViTKeep rate r=0.9, Backbone=DeiT-S2026.0490.58——
PackNet (Reversed)Backbone=ResNet-50, Input Resolution=224x224, Order=Sketch, WikiArt, Flowers, Cars, CUB, Number of Parameters=23.9M, FLOPs=4.11G2019.0690.55——
Ours (Rényi, α = 5.0)Keep rate r=0.7, Backbone=DeiT-S2026.0490.45——
Ours (Shannon)Keep rate r=0.9, Backbone=DeiT-S2026.0490.23——
Ours (Rényi, α = 5.0)Keep rate r=0.8, Backbone=DeiT-S2026.0490.14——