MetaFormer Baselines for Vision

Open paper
Benchmark Model Rank Results
domain-generalization-on-imagenet-aCAFormer-B36 (IN-21K, 384)#4Top-1 accuracy %: 79.5Number of params: 99M
domain-generalization-on-imagenet-aConvFormer-B36 (IN-21K, 384)#7Top-1 accuracy %: 73.5Number of params: 100M
domain-generalization-on-imagenet-aCAFormer-B36 (IN-21K)#8Top-1 accuracy %: 69.4Number of params: 99M
domain-generalization-on-imagenet-aConvFormer-B36 (IN-21K)#11Top-1 accuracy %: 63.3Number of params: 100M
domain-generalization-on-imagenet-aCAFormer-B36 (384)#13Top-1 accuracy %: 61.9Number of params: 99M
domain-generalization-on-imagenet-aConvFormer-B36 (384)#16Top-1 accuracy %: 55.3Number of params: 100M
domain-generalization-on-imagenet-aCAFormer-B36#19Top-1 accuracy %: 48.5Number of params: 99M
domain-generalization-on-imagenet-aConvFormer-B36#22Top-1 accuracy %: 40.1Number of params: 100M
domain-generalization-on-imagenet-cCAFormer-B36 (IN21K, 384)#2mean Corruption Error (mCE): 30.8Number of params: 99M
domain-generalization-on-imagenet-cCAFormer-B36 (IN21K)#5mean Corruption Error (mCE): 31.8
domain-generalization-on-imagenet-cConvFormer-B36 (IN21K)#7mean Corruption Error (mCE): 35.0
domain-generalization-on-imagenet-cCAFormer-B36#18mean Corruption Error (mCE): 42.6
domain-generalization-on-imagenet-cConvFormer-B36#23mean Corruption Error (mCE): 46.3
domain-generalization-on-imagenet-rCAFormer-B36 (IN21K, 384)#4Top-1 Error Rate: 29.6
domain-generalization-on-imagenet-rCAFormer-B36 (IN21K)#6Top-1 Error Rate: 31.7
domain-generalization-on-imagenet-rConvFormer-B36 (IN21K, 384)#10Top-1 Error Rate: 33.5
domain-generalization-on-imagenet-rConvFormer-B36 (IN21K)#12Top-1 Error Rate: 34.7
domain-generalization-on-imagenet-rCAFormer-B36 (384)#20Top-1 Error Rate: 45
domain-generalization-on-imagenet-rCAFormer-B36#22Top-1 Error Rate: 46.1
domain-generalization-on-imagenet-rConvFormer-B36 (384)#23Top-1 Error Rate: 47.8
domain-generalization-on-imagenet-rConvFormer-B36#24Top-1 Error Rate: 48.9
domain-generalization-on-imagenet-sketchCAFormer-B36 (IN21K, 384)#4Top-1 accuracy: 54.5
domain-generalization-on-imagenet-sketchConvFormer-B36 (IN21K, 384)#6Top-1 accuracy: 52.9
domain-generalization-on-imagenet-sketchCAFormer-B36 (IN21K)#7Top-1 accuracy: 52.8
domain-generalization-on-imagenet-sketchConvFormer-B36 (IN21K)#8Top-1 accuracy: 52.7
domain-generalization-on-imagenet-sketchCAFormer-B36#16Top-1 accuracy: 42.5
domain-generalization-on-imagenet-sketchConvFormer-B36#18Top-1 accuracy: 39.5
image-classification-on-imagenetCAFormer-B36 (384 res, 21K)#57Top 1 Accuracy: 88.1%Number of params: 99MGFLOPs: 72.2
image-classification-on-imagenetConvFormer-B36 (384 res, 21K)#78Top 1 Accuracy: 87.6%Number of params: 100MGFLOPs: 66.5
image-classification-on-imagenetCAFormer-M36 (384 res, 21K)#84Top 1 Accuracy: 87.5%Number of params: 56MGFLOPs: 42
image-classification-on-imagenetCAFormer-B36 (224 res, 21K)#91Top 1 Accuracy: 87.4%Number of params: 99MGFLOPs: 23.2
image-classification-on-imagenetConvFormer-B36 (224 res, 21K)#111Top 1 Accuracy: 87.0%Number of params: 100MGFLOPs: 22.6
image-classification-on-imagenetCAFormer-S36 (384 res, 21K)#116Top 1 Accuracy: 86.9%Number of params: 39MGFLOPs: 26.0
image-classification-on-imagenetConvFormer-M36 (384 res, 21K)#117Top 1 Accuracy: 86.9%Number of params: 57MGFLOPs: 37.7
image-classification-on-imagenetCAFormer-M36 (224 res, 21K)#131Top 1 Accuracy: 86.6%Number of params: 56MGFLOPs: 13.2
image-classification-on-imagenetConvFormer-S36 (384 res, 21K)#146Top 1 Accuracy: 86.4%Number of params: 40MGFLOPs: 22.4
image-classification-on-imagenetCAFormer-B36 (384 res)#147Top 1 Accuracy: 86.4%Number of params: 99MGFLOPs: 72.2
image-classification-on-imagenetCAFormer-M36 (384 res)#162Top 1 Accuracy: 86.2%Number of params: 56MGFLOPs: 42.0
image-classification-on-imagenetConvFormer-M36 (224 res, 21K)#169Top 1 Accuracy: 86.1%Number of params: 57MGFLOPs: 12.8
image-classification-on-imagenetCAFormer-S36 (224 res, 21K)#190Top 1 Accuracy: 85.8%Number of params: 39MGFLOPs: 8.0
image-classification-on-imagenetCAFormer-S36 (384 res)#205Top 1 Accuracy: 85.7%Number of params: 39MGFLOPs: 26.0
image-classification-on-imagenetConvFormer-B36 (384 res)#206Top 1 Accuracy: 85.7%Number of params: 100MGFLOPs: 66.5
image-classification-on-imagenetConvFormer-M36 (384 res)#211Top 1 Accuracy: 85.6%Number of params: 57MGFLOPs: 37.7
image-classification-on-imagenetCAFormer-B36 (224 res)#221Top 1 Accuracy: 85.5%Number of params: 99MGFLOPs: 23.2
image-classification-on-imagenetCAFormer-S18 (384 res, 21K)#226Top 1 Accuracy: 85.4%Number of params: 26MGFLOPs: 13.4
image-classification-on-imagenetConvFormer-S36 (224 res, 21K)#227Top 1 Accuracy: 85.4%Number of params: 40MGFLOPs: 7.6
image-classification-on-imagenetConvFormer-S36 (384 res)#228Top 1 Accuracy: 85.4%Number of params: 40MGFLOPs: 22.4
image-classification-on-imagenetCAFormer-M36 (224 res)#244Top 1 Accuracy: 85.2%Number of params: 56MGFLOPs: 13.2
image-classification-on-imagenetCAFormer-S18 (384 res)#259Top 1 Accuracy: 85.0%Number of params: 26MGFLOPs: 13.4
image-classification-on-imagenetConvFormer-S18 (384 res, 21K)#260Top 1 Accuracy: 85.0%Number of params: 27MGFLOPs: 11.6
image-classification-on-imagenetConvFormer-B36 (224 res)#284Top 1 Accuracy: 84.8%Number of params: 100MGFLOPs: 22.6
image-classification-on-imagenetCAFormer-S36 (224 res)#304Top 1 Accuracy: 84.5%Number of params: 39MGFLOPs: 8.0
image-classification-on-imagenetConvFormer-M36 (224 res)#305Top 1 Accuracy: 84.5%Number of params: 57MGFLOPs: 12.8
image-classification-on-imagenetConvFormer-S18 (384 res)#312Top 1 Accuracy: 84.4%Number of params: 27MGFLOPs: 11.6
image-classification-on-imagenetCAFormer-S18 (224 res, 21K)#345Top 1 Accuracy: 84.1%Number of params: 26MGFLOPs: 4.1
image-classification-on-imagenetConvFormer-S36 (224 res)#346Top 1 Accuracy: 84.1%Number of params: 40MGFLOPs: 7.6
image-classification-on-imagenetConvFormer-S18 (224 res, 21K)#387Top 1 Accuracy: 83.7%Number of params: 27MGFLOPs: 3.9
image-classification-on-imagenetCAFormer-S18 (224 res)#404Top 1 Accuracy: 83.6%Number of params: 26MGFLOPs: 4.1
image-classification-on-imagenetConvFormer-S18 (224 res)#465Top 1 Accuracy: 83.0%Number of params: 27MGFLOPs: 3.9

Rank counts only results with a code link.