Image Classification on ImageNet 2012 (val)
88.5Top-1 AccuracyFixEfficientNet-L2
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| FixEfficientNet-L2Architecture=L2, #params=480M, train res=475, test res=6002020.03 | 88.5 | — | — | 98.7 | — | |
| EfficientNet-L2Architecture=L2, #params=480M, train res=475, test res=8002020.03 | 88.4 | — | — | 98.7 | — | |
| FixEfficientNet-B7Architecture=B7, #params=66M, train res=600, test res=6322020.03 | 87.1 | — | — | 98.2 | — | |
| EfficientNet-B7Architecture=B7, #params=66M, train res=600, test res=6002020.03 | 86.9 | — | — | 98.1 | — | |
| FixEfficientNet-B6Architecture=B6, #params=43M, train res=528, test res=6802020.03 | 86.7 | — | — | 98 | — | |
| FixEfficientNet-B5Architecture=B5, #params=30M, train res=456, test res=5762020.03 | 86.4 | — | — | 97.9 | — | |
| EfficientNet-B6Architecture=B6, #params=43M, train res=528, test res=5282020.03 | 86.4 | — | — | 97.9 | — | |
| EfficientNet-B5Architecture=B5, #params=30M, train res=456, test res=4562020.03 | 86.1 | — | — | 97.8 | — | |
| FixEfficientNet-B4Architecture=B4, #params=19M, train res=380, test res=4722020.03 | 85.9 | — | — | 97.7 | — | |
| MaxUp + CutMixBackbone=Fix-EfficientNet-B8, Model Size=87.42M, FLOPs=101.79G, Training Epochs (Fine-tuning)=52020.02 | 85.8 | — | — | — | — | |
| FixEfficientNetModel Variant=B8, #params=87.4M, train res=672, test res=8002020.03 | 85.7 | — | — | 97.6 | — | |
| CutMixBackbone=Fix-EfficientNet-B8, Model Size=87.42M, FLOPs=101.79G, Training Epochs (Fine-tuning)=52020.02 | 85.57 | — | — | — | — | |
| EfficientNet [9]Model Variant=B8, #params=87.4M, train res=672, test res=6722020.03 | 85.5 | — | — | 97.3 | — | |
| MaxUp + CutMixBackbone=EfficientNet-B7, Model Size=66.35M, FLOPs=38.20G, Training Epochs (Fine-tuning)=52020.02 | 85.45 | — | — | — | — | |
| EfficientNet-B4Architecture=B4, #params=19M, train res=380, test res=3802020.03 | 85.3 | — | — | 97.5 | — | |
| FixEfficientNetModel Variant=B7, #params=66M, train res=600, test res=6322020.03 | 85.3 | — | — | 97.4 | — | |
| CutMixBackbone=EfficientNet-B7, Model Size=66.35M, FLOPs=38.20G, Training Epochs (Fine-tuning)=52020.02 | 85.22 | — | — | — | — | |
| EfficientNet [9]Model Variant=B7, #params=66M, train res=600, test res=6002020.03 | 85.2 | — | — | 97.2 | — | |
| FixEfficientNet-B3Architecture=B3, #params=12M, train res=300, test res=4722020.03 | 85 | — | — | 97.4 | — | |
| FixEfficientNetModel Variant=B6, #params=43M, train res=528, test res=5762020.03 | 84.9 | — | — | 97.3 | — | |
| EfficientNet [9]Model Variant=B6, #params=43M, train res=528, test res=5282020.03 | 84.8 | — | — | 97.1 | — | |
| FixEfficientNetModel Variant=B5, #params=30M, train res=456, test res=5762020.03 | 84.7 | — | — | 97.2 | — | |
| EfficientNet-B7MParams=66.7, GFLOPs=37.0, Crop=600, Batch=4096, Epochs=3502020.11 | 84.4 | — | — | 97.1 | — | |
| EfficientNet [9]Model Variant=B5, #params=30M, train res=456, test res=4562020.03 | 84.3 | — | — | 97 | — | |
| EfficientNet-B6MParams=43.0, GFLOPs=19.0, Crop=528, Batch=4096, Epochs=3502020.11 | 84.2 | — | — | 96.8 | — | |
| EfficientNet-B3Architecture=B3, #params=12M, train res=300, test res=3002020.03 | 84.1 | — | — | 96.9 | — | |
| FixEfficientNetModel Variant=B4, #params=19M, train res=380, test res=5122020.03 | 84 | — | — | 97 | — | |
| ResNeSt-200MParams=70.4, GFLOPs=35.6, Crop=320, Batch=2048, Epochs=2702020.11 | 83.88 | — | — | — | — | |
| FixEfficientNet-B2Architecture=B2, #params=9.2M, train res=260, test res=4202020.03 | 83.6 | — | — | 96.9 | — | |
| SE-ResNeXt-101, 64×4d, S = 2MParams=98.0, GFLOPs=38.2, Crop=320, Batch=128, Epochs=350, S=22020.11 | 83.6 | — | — | 96.69 | — | |
| EfficientNet [9]Model Variant=B4, #params=19M, train res=380, test res=3802020.03 | 83.3 | — | — | 96.4 | — | |
| SENet-154 + MultiGrainMParams=115.0, GFLOPs=83.1, Crop=450, Batch=512, Epochs=1202020.11 | 83.1 | — | — | 96.5 | — | |
| AmoebaNet-CN=6, F=228, MParams=155.3, GFLOPs=41.1, Crop=331, Batch=3200, Epochs=1002020.11 | 83.1 | — | — | 96.3 | — | |
| FixEfficientNetModel Variant=B3, #params=12M, train res=300, test res=4722020.03 | 83 | — | — | 96.4 | — | |
| PNASNet-5N=4, F=216, MParams=86.1, GFLOPs=25.0, Crop=331, Batch=1600, Epochs=3122020.11 | 82.9 | — | — | 96.2 | — | |
| EfficientNet-B7, S = 4MParams=70.9, GFLOPs=12.6, Crop=320, Batch=256, Epochs=120, S=42020.11 | 82.75 | — | — | 96.22 | — | |
| EfficientNet-B7, S = 2MParams=68.2, GFLOPs=10.5, Crop=320, Batch=256, Epochs=120, S=22020.11 | 82.74 | — | — | 96.3 | — | |
| SENet-154MParams=115.0, GFLOPs=42.3, Crop=320, Batch=1024, Epochs=1002020.11 | 82.72 | — | — | 96.21 | — | |
| FixEfficientNet-B1Architecture=B1, #params=7.8M, train res=240, test res=3842020.03 | 82.6 | — | — | 96.5 | — | |
| EfficientNet-B2Architecture=B2, #params=9.2M, train res=260, test res=2602020.03 | 82.4 | — | — | 96.3 | — | |
| ResNeXt-101, 64×4d, S = 2MParams=88.6, GFLOPs=18.8, Crop=224, Batch=256, Epochs=120, S=22020.11 | 82.13 | — | — | 95.98 | — | |
| FixEfficientNetModel Variant=B2, #params=9.2M, train res=260, test res=4202020.03 | 82 | — | — | 96 | — | |
| EfficientNet [9]Model Variant=B3, #params=12M, train res=300, test res=3002020.03 | 81.9 | — | — | 95.6 | — | |
| EfficientNet-B7 (re_impl.)MParams=66.7, GFLOPs=10.6, Crop=320, Batch=256, Epochs=1202020.11 | 81.83 | — | — | 95.78 | — | |
| ResNeXt-101, 64×4d (re_impl.)MParams=83.6, GFLOPs=16.9, Crop=224, Batch=256, Epochs=1202020.11 | 81.57 | — | — | 95.73 | — | |
| EfficientNet-B1Architecture=B1, #params=7.8M, train res=240, test res=2402020.03 | 81.5 | — | — | 95.8 | — | |
| WRN-50-3, S = 2MParams=138.0, GFLOPs=25.6, Crop=224, Batch=256, Epochs=120, S=22020.11 | 81.42 | — | — | 95.62 | — | |
| FixEfficientNetModel Variant=B1, #params=7.8M, train res=240, test res=3842020.03 | 81.3 | — | — | 95.7 | — | |
| WRN-50-3 (re_impl.)MParams=135.0, GFLOPs=23.8, Crop=224, Batch=256, Epochs=1202020.11 | 80.74 | — | — | 95.4 | — | |
| WRN-50-2 (re_impl.)MParams=68.9, GFLOPs=12.8, Crop=224, Batch=256, Epochs=1202020.11 | 80.66 | — | — | 95.16 | — | |
| EfficientNet [9]Model Variant=B2, #params=9.2M, train res=260, test res=2602020.03 | 80.5 | — | — | 95 | — | |
| NSGANetV2-xlType=auto, Search Cost (GPU days)=1†, #Params=8.7M, #MAdds=593M, CPU Latency (ms)=16.7, GPU Latency (ms)=732020.07 | 80.4 | — | — | 95.2 | — | |
| MaxUp + CutMixBackbone=ResNet-101, Model Size=44.55M, FLOPs=7.85G, Training Epochs (Fine-tuning)=202020.02 | 80.26 | — | — | — | — | |
| FixEfficientNet-B0Architecture=B0, #params=5.3M, train res=224, test res=3202020.03 | 80.2 | — | — | 95.4 | — | |
| ResNet-200 + AutoAugmentMParams=64.8, GFLOPs=16.4, Crop=224, Batch=4096, Epochs=2702020.11 | 80 | — | — | 95 | — | |
| CutMixBackbone=ResNet-101, Model Size=44.55M, FLOPs=7.85G, Training Epochs (Fine-tuning)=202020.02 | 79.83 | — | — | — | — | |
| DeiT (dense)Architecture=DeiT-Small, Sparsity (s%)=0.002025.04 | 79.8 | — | — | — | — | |
| CBAM-R50 + CGCTraining Setting=Advanced, Param=28.12M, ΔMFLOPs=212019.10 | 79.74 | — | — | 94.83 | — | |
| WRN-50-2, S = 2MParams=51.4, GFLOPs=10.9, Crop=224, Batch=256, Epochs=120, S=22020.11 | 79.64 | — | — | 94.82 | — | |
| EfficientNet [9]Model Variant=B1, #params=7.8M, train res=240, test res=2402020.03 | 79.6 | — | — | 94.3 | — | |
| ResNeXt-101, 64×4dMParams=83.6, GFLOPs=16.9, Crop=224, Batch=256, Epochs=1202020.11 | 79.6 | — | — | 94.7 | — | |
| R50 + CGCTraining Setting=Advanced, Param=25.59M, ΔMFLOPs=62019.10 | 79.54 | — | — | 94.78 | — | |
| FixEfficientNetModel Variant=B0, #params=5.3M, train res=224, test res=3202020.03 | 79.3 | — | — | 94.6 | — | |
| NSGANetV2-lType=auto, Search Cost (GPU days)=1†, #Params=8.0M, #MAdds=400M, CPU Latency (ms)=12.9, GPU Latency (ms)=522020.07 | 79.1 | — | — | 94.5 | — | |
| MixNet-LType=auto, Parameters=7.3M, FLOPS=565M, multiplier=1.3x2019.07 | 78.9 | — | — | 94.2 | — | |
| MixNet-LType=auto, #Params=7.3M, #MAdds=565M, CPU Latency (ms)=29.4, GPU Latency (ms)=1052020.07 | 78.9 | — | — | 94.2 | — | |
| DCNv2-R50Training Setting=Advanced, Param=27.4M, ΔMFLOPs=2002019.10 | 78.89 | — | — | 94.6 | — | |
| DIANetBackbone=ResNet152, #P(M)=65.85, reduction ratio (r)=102019.05 | 78.87 | — | — | — | — | |
| CBAM-R50Training Setting=Advanced, Param=28.09M, ΔMFLOPs=152019.10 | 78.86 | — | — | 94.58 | — | |
| EfficientNet-B1Type=auto, #Params=7.8M, #MAdds=700M, CPU Latency (ms)=21.5, GPU Latency (ms)=782020.07 | 78.8 | — | — | 94.4 | — | |
| EfficientNet-B0Architecture=B0, #params=5.3M, train res=224, test res=2242020.03 | 78.8 | — | — | 94.5 | — | |
| SE-R50Training Setting=Advanced, Param=28.09M, ΔMFLOPs=82019.10 | 78.79 | — | — | 94.52 | — | |
| R50+ GloReTraining Setting=Default, Param=30.5M, ΔMFLOPs=12002019.10 | 78.4 | — | — | — | — | |
| SENetBackbone=ResNet152, #P(M)=66.822019.05 | 78.36 | — | — | — | — | |
| DIANetBackbone=ResNext50,32x4, #P(M)=27.83, reduction ratio (r)=202019.05 | 78.32 | — | — | — | — | |
| NSGANetV2-mType=auto, Search Cost (GPU days)=1†, #Params=7.7M, #MAdds=312M, CPU Latency (ms)=11.4, GPU Latency (ms)=372020.07 | 78.3 | — | — | 94.1 | — | |
| DCNv2-R50Training Setting=Default, Param=27.4M, ΔMFLOPs=2002019.10 | 78.2 | — | — | 94 | — | |
| R50Training Setting=Advanced, Param=25.56M, ΔMFLOPs=02019.10 | 78.13 | — | — | 94.06 | — | |
| WRN-50-2MParams=68.9, GFLOPs=12.8, Crop=224, Batch=256, Epochs=1202020.11 | 78.1 | — | — | 93.97 | — | |
| SENetBackbone=ResNext50,32x4, #P(M)=27.562019.05 | 78.04 | — | — | — | — | |
| CAPArchitecture=DeiT-Small, Sparsity (s%)=80.002025.04 | 78 | — | — | — | — | |
| HyperfluxArchitecture=DeiT-Small, Sparsity (s%)=80.002025.04 | 77.8 | — | — | — | — | |
| AutoNL-L (ours)#Params=5.6M, Flops=353M2020.04 | 77.7 | — | — | 93.7 | — | |
| GC-R50Training Setting=Default, Param=28.08M, ΔMFLOPs=1002019.10 | 77.7 | — | — | 93.66 | — | |
| CBAM-R50 + CGCTraining Setting=Default, Param=28.12M, ΔMFLOPs=212019.10 | 77.68 | — | — | 93.68 | — | |
| EfficientNet [9]Model Variant=B0, #params=5.3M, train res=224, test res=2242020.03 | 77.6 | — | — | 93.3 | — | |
| ResNet152Backbone=ResNet152, #P(M)=60.272019.05 | 77.58 | — | — | — | — | |
| R50 + CGCTraining Setting=Default, Param=25.59M, ΔMFLOPs=62019.10 | 77.48 | — | — | 93.81 | — | |
| NSGANetV2-sType=auto, Search Cost (GPU days)=1†, #Params=6.1M, #MAdds=225M, CPU Latency (ms)=9.1, GPU Latency (ms)=302020.07 | 77.4 | — | — | 93.5 | — | |
| CBAM-R50Training Setting=Default, Param=28.09M, ΔMFLOPs=152019.10 | 77.34 | — | — | 93.69 | — | |
| GMArchitecture=DeiT-Small, Sparsity (s%)=80.002025.04 | 77.3 | — | — | — | — | |
| DIANetBackbone=ResNet50, #P(M)=28.38, reduction ratio (r)=202019.05 | 77.24 | — | — | — | — | |
| DCNv2-R50Training Setting=Default, Param=27.4M, ΔMFLOPs=2002019.10 | 77.21 | — | — | 93.69 | — | |
| AtomNAS-C+Type=auto, Search Cost (GPU days)=1†, #Params=5.5M, #MAdds=329M2020.07 | 77.2 | — | — | 93.5 | — | |
| ResNext50,32x4Backbone=ResNext50,32x4, #P(M)=25.032019.05 | 77.19 | — | — | — | — | |
| SE-R50Training Setting=Default, Param=28.09M, ΔMFLOPs=82019.10 | 77.18 | — | — | 93.67 | — | |
| MaxUp + CutMixBackbone=ProxylessNet-Mobile x1.4, Model Size=6.86M, FLOPs=603M, Training Epochs (Fine-tuning)=202020.02 | 77.17 | — | — | — | — | |
| SK-R50Training Setting=Default, Param=37.25M, ΔMFLOPs=18372019.10 | 77.15 | — | — | 93.54 | — | |
| ResNet-50Params=25.6M, s(%)=0.00, Ftest=1.00×, Ftrain=1.00×2025.04 | 77.01 | — | — | — | — | |
| ResNet-153Type=manual, Parameters=60M, FLOPS=11B2019.07 | 77 | — | — | 93.3 | — |