Image Classification on ImageNet (val) (Top-1 and Top-5 Accuracy)
83.9Top-1 AccuracyAmoebaNet-A
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| AmoebaNet-AN=6, F=448, # Parameters=469M, # Multiply-Adds=104B, evolution-based=true2018.02 | 83.9 | 96.6 | — | |
| ViTAE-B-StageType=Transformer, Params (M)=48.5, MACs (G)=13.8, Input Size=2242021.06 | 83.6 | 96.4 | — | |
| AutoAugmentModel=AmoebaNet-C (6,228)2018.05 | 83.5 | 96.5 | — | |
| Conformer-SType=Transformer, Params (M)=37.7, MACs (G)=10.6, Input Size=2242021.06 | 83.4 | — | — | |
| Inception Pre-processingModel=AmoebaNet-C (6,228)2018.05 | 83.1 | 96.1 | — | |
| Swin-SType=Transformer, Params (M)=50, MACs (G)=17.4, Input Size=2242021.06 | 83 | — | — | |
| PNASNet-5# Parameters=86.1M, # Multiply-Adds=25.0B2018.02 | 82.9 | 96.2 | — | |
| EfficientNet-B4Type=CNN, Params (M)=19.3, MACs (G)=8.4, Input Size=3802021.06 | 82.9 | 96.4 | — | |
| AmoebaNet-AN=6, F=190, # Parameters=86.7M, # Multiply-Adds=23.1B, evolution-based=true2018.02 | 82.8 | 96.1 | — | |
| AutoAugmentModel=AmoebaNet-B (6,190)2018.05 | 82.8 | 96.2 | — | |
| TNT-BType=Transformer, Params (M)=65.6, MACs (G)=28.2, Input Size=2242021.06 | 82.8 | 96.3 | — | |
| NASNet-A# Parameters=88.9M, # Multiply-Adds=23.8B2018.02 | 82.7 | 96.2 | — | |
| CvT-21Type=Transformer, Params (M)=32, MACs (G)=14.2, Input Size=2242021.06 | 82.5 | — | — | |
| ConViT-B+Type=Transformer, Params (M)=152, MACs (G)=60, Input Size=2242021.06 | 82.5 | — | — | |
| ConViT-BType=Transformer, Params (M)=86, MACs (G)=34, Input Size=2242021.06 | 82.4 | — | — | |
| Inception Pre-processingModel=AmoebaNet-B (6,190)2018.05 | 82.2 | 96 | — | |
| ViTAE-S-StageType=Transformer, Params (M)=19.2, MACs (G)=6, Input Size=2242021.06 | 82.2 | 96 | — | |
| ConViT-S+Type=Transformer, Params (M)=48, MACs (G)=20, Input Size=2242021.06 | 82.2 | — | — | |
| CeiT-SType=Transformer, Params (M)=24.2, MACs (G)=9, Input Size=2242021.06 | 82 | 95.9 | — | |
| ViTAE-SType=Transformer, Params (M)=23.6, MACs (G)=5.6, Input Size=2242021.06 | 82 | 95.9 | — | |
| PiT-BType=Transformer, Params (M)=73.8, MACs (G)=25, Input Size=2242021.06 | 82 | — | — | |
| T2T-ViT-19Type=Transformer, Params (M)=39.2, MACs (G)=8.9, Input Size=2242021.06 | 81.9 | 95.7 | — | |
| DeiT-BType=Transformer, Params (M)=86.6, MACs (G)=34.6, Input Size=2242021.06 | 81.8 | 95.6 | — | |
| ConT-BType=Transformer, Params (M)=39.6, MACs (G)=12.8, Input Size=2242021.06 | 81.8 | — | — | |
| RegNetY-8GFType=CNN, Params (M)=39.2, MACs (G)=16, Input Size=2242021.06 | 81.7 | — | — | |
| Twins-SVT-SType=Transformer, Params (M)=24, MACs (G)=5.6, Input Size=2242021.06 | 81.7 | — | — | |
| PVT-LType=Transformer, Params (M)=61.4, MACs (G)=19.6, Input Size=2242021.06 | 81.7 | — | — | |
| CvT-13Type=Transformer, Params (M)=20, MACs (G)=9, Input Size=2242021.06 | 81.6 | — | — | |
| Dual-Path-Net-131# Parameters=79.5M, # Multiply-Adds=32.0B2018.02 | 81.5 | 95.8 | — | |
| T2T-ViT-14Type=Transformer, Params (M)=21.5, MACs (G)=5.2, Input Size=2242021.06 | 81.5 | 95.7 | — | |
| PolyNet# Parameters=92.0M, # Multiply-Adds=34.7B2018.02 | 81.3 | 95.8 | — | |
| Conformer-TiType=Transformer, Params (M)=23.5, MACs (G)=5.2, Input Size=2242021.06 | 81.3 | — | — | |
| Swin-TType=Transformer, Params (M)=29, MACs (G)=9, Input Size=2242021.06 | 81.3 | — | — | |
| ConViT-SType=Transformer, Params (M)=27, MACs (G)=10.8, Input Size=2242021.06 | 81.3 | — | — | |
| TNT-SType=Transformer, Params (M)=23.8, MACs (G)=10.4, Input Size=2242021.06 | 81.3 | 95.6 | — | |
| DeiT-SType=Transformer, Params (M)=22.1, MACs (G)=9.8, Input Size=224, distilled=true2021.06 | 81.2 | 95.4 | — | |
| Twins-PCPVT-SType=Transformer, Params (M)=24.1, MACs (G)=7.4, Input Size=2242021.06 | 81.2 | — | — | |
| PVT-MType=Transformer, Params (M)=44.2, MACs (G)=13.2, Input Size=2242021.06 | 81.2 | — | — | |
| ViTAE-13MType=Transformer, Params (M)=13.2, MACs (G)=3.4, Input Size=2242021.06 | 81 | 95.4 | — | |
| CrossViT-SType=Transformer, Params (M)=26.7, MACs (G)=11.2, Input Size=2242021.06 | 81 | — | — | |
| ResNeXt-101# Parameters=83.6M, # Multiply-Adds=31.5B2018.02 | 80.9 | 95.6 | — | |
| PiT-SType=Transformer, Params (M)=23.5, MACs (G)=4.8, Input Size=2242021.06 | 80.9 | — | — | |
| Incep-ResNet V2# Parameters=55.8M, # Multiply-Adds=13.2B2018.02 | 80.4 | 95.3 | — | |
| ConT-MType=Transformer, Params (M)=19.2, MACs (G)=6.2, Input Size=2242021.06 | 80.2 | — | — | |
| AutoAugmentModel=ResNet-2002018.05 | 80 | 95 | — | |
| RegNetY-4GFType=CNN, Params (M)=20.6, MACs (G)=8, Input Size=2242021.06 | 80 | — | — | |
| DeiT-SType=Transformer, Params (M)=22.1, MACs (G)=9.8, Input Size=224, distilled=false2021.06 | 79.9 | 95 | — | |
| PVT-SType=Transformer, Params (M)=24.5, MACs (G)=7.6, Input Size=2242021.06 | 79.8 | — | — | |
| Hierarchical# Parameters=64M, evolution-based=true2018.02 | 79.7 | 94.8 | — | |
| LIP-ResNet-101#Params=42.9M, FLOPS=9.06G2019.08 | 79.33 | 94.6 | — | |
| ResNet-152Type=CNN, Params (M)=60.2, MACs (G)=22.6, Input Size=2242021.06 | 78.9 | 94.4 | — | |
| Inception Pre-processingModel=ResNet-2002018.05 | 78.5 | 94.2 | — | |
| LR-Net-101# params=42.0M, FLOPS=8.0G2019.04 | 78.5 | 94.3 | — | |
| ResNet-152#Params=60.2M, FLOPS=11.58G, recipe=similar recipe2019.08 | 78.49 | 94.22 | — | |
| ResNet-101Type=CNN, Params (M)=44.5, MACs (G)=15.2, Input Size=2242021.06 | 78.3 | 94.1 | — | |
| Res2NeXt-50#Parameter=24.27M, BFLOPs=8.4 x 10^9, Evaluation Protocol=10 crop2019.11 | 78.2 | 93.9 | — | |
| CSPResNeXt-50#Parameter=20.50M, BFLOPs=7.9 x 10^9, Evaluation Protocol=10 crop2019.11 | 78.2 | 94.3 | — | |
| LocalViT-PVTType=Transformer, Params (M)=13.5, MACs (G)=9.6, Input Size=2242021.06 | 78.2 | 94.2 | — | |
| LIP-ResNet-50#Params=23.9M, FLOPS=5.33G2019.08 | 78.19 | 93.96 | — | |
| PiT-XSType=Transformer, Params (M)=10.6, MACs (G)=2.8, Input Size=2242021.06 | 78.1 | — | — | |
| INOLMLBackbone=ResNet-50, Model cotraining or ensembling=true, Self-supervised pre-training=false2022.11 | 78.1 | 92.9 | — | |
| Res2Net-50#Parameter=25.29M, BFLOPs=8.4 x 10^9, Evaluation Protocol=10 crop2019.11 | 78 | 93.8 | — | |
| ResNet-101#Params=44.5M, FLOPS=7.85G2019.08 | 77.98 | 93.98 | — | |
| ResNet-101# params=44.4M, FLOPS=8.0G2019.04 | 77.9 | 94 | — | |
| CSPResNeXt-50#Parameter=20.50M, BFLOPs=7.93 (-22%)2019.11 | 77.9 | 94 | — | |
| DenseNet-201-Elastic#Parameter=19.48M, BFLOPs=8.772019.11 | 77.9 | 94 | — | |
| CSPDenseNet-201-Elastic#Parameter=20.17M, BFLOPs=7.13 (-19%)2019.11 | 77.9 | 94 | — | |
| ViTAE-6MType=Transformer, Params (M)=6.5, MACs (G)=2, Input Size=2242021.06 | 77.9 | 94.1 | — | |
| ViT-B/16Type=Transformer, Params (M)=86.5, MACs (G)=18.7, Input Size=3842021.06 | 77.9 | — | — | |
| ResNeXt-50#Parameter=22.19M, BFLOPs=10.112019.11 | 77.8 | 94.2 | — | |
| HarDNet-138s#Parameter=35.5M, BFLOPs=13.42019.11 | 77.8 | — | — | |
| DenseNet-264-32#Parameter=27.21M, BFLOPs=11.032019.11 | 77.8 | 93.9 | — | |
| ResNet-152#Parameter=60.2M, BFLOPs=22.62019.11 | 77.8 | 93.6 | — | |
| SupervisedArchitecture=ResNet-101, Evaluation Protocol=Linear evaluation, Crop Type=single centred crop2019.11 | 77.8 | 93.8 | — | |
| AutoAugmentModel=ResNet-502018.05 | 77.6 | 93.8 | — | |
| Dynamic LossBackbone=ResNet-50, Model cotraining or ensembling=false, Self-supervised pre-training=CLIP2022.11 | 77.43 | 93.4 | — | |
| SVD-PadéBackbone=ResNet-502022.05 | 77.33 | 93.49 | — | |
| LR-Net-50# params=23.3M, FLOPS=4.3G2019.04 | 77.3 | 93.6 | — | |
| iSQRT-COVBackbone=ResNet-502022.05 | 77.19 | 93.4 | — | |
| EfficientNet-B0Type=CNN, Params (M)=5.3, MACs (G)=0.8, Input Size=2242021.06 | 77.1 | 93.3 | — | |
| EfficientNet-B0Params (M)=5.3, Multiply-Adds (M)=390, Search Method=RL2020.06 | 77.1 | 93.2 | — | |
| MPN-COVBackbone=ResNet-502022.05 | 77.07 | 93.25 | — | |
| FaMUSBackbone=IRV2, Model cotraining or ensembling=true, Self-supervised pre-training=false2022.11 | 77 | 92.76 | — | |
| ViTAE-T-StageType=Transformer, Params (M)=4.8, MACs (G)=2.3, Input Size=2242021.06 | 76.8 | 93.5 | — | |
| MnasNet-A3Type=auto, #Params=5.2M, #Mult-Adds=403M, Inference Latency=103ms2018.07 | 76.7 | 93.3 | — | |
| ResNet-50Type=CNN, Params (M)=25.6, MACs (G)=7.6, Input Size=2242021.06 | 76.7 | 93.3 | — | |
| ConViT-Ti+Type=Transformer, Params (M)=10, MACs (G)=4, Input Size=2242021.06 | 76.7 | — | — | |
| LIP-DenseNet-BC-121#Params=8.7M, FLOPS=4.13G2019.08 | 76.64 | 93.16 | — | |
| MobileNet-V3Params (M)=7.5, Multiply-Adds (M)=356, Search Method=RL2020.06 | 76.6 | — | — | |
| YOLO-Master-cls-NSize=2242025.12 | 76.6 | 93.4 | — | |
| ViT-L/16Type=Transformer, Params (M)=304.3, MACs (G)=65.8, Input Size=3842021.06 | 76.5 | — | — | |
| SupervisedEvaluation Protocol=Linear evaluation2020.11 | 76.5 | — | — | |
| ResNet-50#Params=25.6M, FLOPS=4.12G2019.08 | 76.4 | 93.15 | — | |
| CeiT-TType=Transformer, Params (M)=6.4, MACs (G)=2.4, Input Size=2242021.06 | 76.4 | 93.4 | — | |
| Inception Pre-processingModel=ResNet-502018.05 | 76.3 | 93.1 | — | |
| ResNet-50# params=25.5M, FLOPS=4.3G2019.04 | 76.3 | 93.2 | — | |
| SupervisedArchitecture=ResNet-50, Evaluation Protocol=Linear evaluation, Crop Type=single centred crop2019.11 | 76.3 | 93.1 | — | |
| Teacher (ResNet-50)Teacher Architecture=ResNet-50, Training Strategy=B12022.05 | 76.16 | 92.86 | — | |
| Teacher (ResNet-50)Architecture=ResNet-502024.03 | 76.16 | 92.86 | — | |
| XNASParams (M)=5.2, Search Cost (GPU days)=0.3, Search Method=gradient2020.06 | 76 | — | — |