Incorporating Convolution Designs into Visual Transformers

Open paper
Benchmark Model Rank Results
image-classification-on-cifar-10CeiT-S (384 finetune resolution)#16Percentage correct: 99.1
image-classification-on-cifar-10CeiT-S#23Percentage correct: 99
image-classification-on-cifar-10CeiT-T#40Percentage correct: 98.5
image-classification-on-cifar-100CeiT-S#16Percentage correct: 91.8
image-classification-on-cifar-100CeiT-S (384 finetune resolution)#17Percentage correct: 91.8
image-classification-on-cifar-100CeiT-T#31Percentage correct: 89.4
image-classification-on-cifar-100CeiT-T (384 finetune resolution)#40Percentage correct: 88
image-classification-on-flowers-102CeiT-S (384 finetune resolution)#20Accuracy: 98.6
image-classification-on-flowers-102CeiT-S#26Accuracy: 98.2
image-classification-on-flowers-102CeiT-T (384 finetune resolution)#33Accuracy: 97.8
image-classification-on-flowers-102CeiT-T#38Accuracy: 96.9
image-classification-on-imagenetCeiT-S (384 finetune res)#425Top 1 Accuracy: 83.3%Number of params: 24.2MGFLOPs: 12.9
image-classification-on-imagenetCeiT-S#561Top 1 Accuracy: 82%GFLOPs: 4.5
image-classification-on-imagenetCeiT-T (384 finetune res)#761Top 1 Accuracy: 78.8%GFLOPs: 3.6
image-classification-on-imagenetCeiT-T#857Top 1 Accuracy: 76.4%Number of params: 6.4MGFLOPs: 1.2
image-classification-on-imagenet-realCeiT-S (384 finetune res)#28Accuracy: 88.1%
image-classification-on-imagenet-realCeiT-S#34Accuracy: 87.3%
image-classification-on-imagenet-realCeiT-T#46Accuracy: 83.6%
image-classification-on-inaturalist-2018CeiT-S (384 finetune resolution)#16Top-1 Accuracy: 79.4%
image-classification-on-inaturalist-2018CeiT-S#29Top-1 Accuracy: 73.3%
image-classification-on-inaturalist-2018CeiT-T (384 finetune resolution)#32Top-1 Accuracy: 72.2%
image-classification-on-inaturalist-2018CeiT-T#43Top-1 Accuracy: 64.3%
image-classification-on-inaturalist-2019CeiT-S (384 finetune resolution)#8Top-1 Accuracy: 82.7
image-classification-on-inaturalist-2019CeiT-S#11Top-1 Accuracy: 78.9
image-classification-on-inaturalist-2019CeiT-T (384 finetune resolution)#12Top-1 Accuracy: 77.9
image-classification-on-inaturalist-2019CeiT-T#15Top-1 Accuracy: 72.8
image-classification-on-oxford-iiit-pets-1CeiT-S (384 finetune resolution)#1Accuracy: 94.9
image-classification-on-oxford-iiit-pets-1CeiT-S#3Accuracy: 94.6
image-classification-on-oxford-iiit-pets-1CeiT-T (384 finetune resolution)#4Accuracy: 94.5
image-classification-on-oxford-iiit-pets-1CeiT-T#5Accuracy: 93.8
image-classification-on-stanford-carsCeiT-S (384 finetune resolution)#5Accuracy: 94.1
image-classification-on-stanford-carsCeiT-S#8Accuracy: 93.2
image-classification-on-stanford-carsCeiT-T (384 finetune resolution)#10Accuracy: 93
image-classification-on-stanford-carsCeiT-T#12Accuracy: 90.5

Rank counts only results with a code link.