Image Classification on ImageNet-1k 1.0 (val) (Top-1 Accuracy, GPU Throughput, CPU/ARM Latency)
83.8Top-1 AccuracyConvNeXt-B
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| ConvNeXt-BParams (M)=88.6, FLOPs (G)=15.38, Resolution=224x2242023.03 | 83.8 | 253 | 257.1 | — | |
| Swin-BParams (M)=87.8, FLOPs (G)=15.47, Resolution=224x2242023.03 | 83.5 | 237 | 349.2 | — | |
| FasterNet-LParams (M)=93.5, FLOPs (G)=15.52, Resolution=224x2242023.03 | 83.5 | 323 | 219.5 | — | |
| ConvNeXt-SParams (M)=50.2, FLOPs (G)=8.71, Resolution=224x2242023.03 | 83.1 | 377 | 153.2 | 3,484 | |
| Swin-SParams (M)=49.6, FLOPs (G)=8.77, Resolution=224x2242023.03 | 83 | 348 | 224.2 | 2,613 | |
| FasterNet-MParams (M)=53.5, FLOPs (G)=8.74, Resolution=224x2242023.03 | 83 | 500 | 129.5 | 2,092 | |
| PoolFormer-M48Params (M)=73.5, FLOPs (G)=11.59, Resolution=224x2242023.03 | 82.5 | 242 | 281.8 | — | |
| ConvNeXt-TParams (M)=28.6, FLOPs (G)=4.47, Resolution=224x2242023.03 | 82.1 | 657 | 86.3 | 1,889 | |
| PoolFormer-M36Params (M)=56.2, FLOPs (G)=8.8, Resolution=224x2242023.03 | 82.1 | 320 | 215 | 2,979 | |
| PVT-LargeParams (M)=61.4, FLOPs (G)=9.85, Resolution=224x2242023.03 | 81.7 | 306 | 203.4 | 3,101 | |
| CycleMLP-B2Params (M)=26.8, FLOPs (G)=3.9, Resolution=224x2242023.03 | 81.6 | 528 | 186.3 | 1,502 | |
| PoolFormer-S36Params (M)=30.9, FLOPs (G)=5, Resolution=224x2242023.03 | 81.4 | 507 | 138 | 1,860 | |
| Swin-TParams (M)=28.3, FLOPs (G)=4.51, Resolution=224x2242023.03 | 81.3 | 609 | 122.2 | 1,424 | |
| FasterNet-SParams (M)=31.1, FLOPs (G)=4.56, Resolution=224x2242023.03 | 81.3 | 1,029 | 71.2 | 1,103 | |
| PVT-MediumParams (M)=44.2, FLOPs (G)=6.69, Resolution=224x2242023.03 | 81.2 | 438 | 143.6 | 2,142 | |
| PoolFormer-S24Params (M)=21.4, FLOPs (G)=3.41, Resolution=224x2242023.03 | 80.3 | 748 | 92.8 | 1,261 | |
| PVT-SmallParams (M)=24.5, FLOPs (G)=3.83, Resolution=224x2242023.03 | 79.8 | 689 | 89.6 | 1,345 | |
| EdgeNeXt-SParams (M)=5.6, FLOPs (G)=1.26, Resolution=256x2562023.03 | 79.4 | 1,128 | 39.2 | 743 | |
| CycleMLP-B1Params (M)=15.2, FLOPs (G)=2.1, Resolution=224x2242023.03 | 79.1 | 865 | 116.1 | 892 | |
| FasterNet-T2Params (M)=15, FLOPs (G)=1.91, Resolution=224x2242023.03 | 78.9 | 1,991 | 33.5 | 497 | |
| ResNet50Params (M)=25.6, FLOPs (G)=4.11, Resolution=224x2242023.03 | 78.8 | 959 | 73 | 1,131 | |
| MobileViT-SParams (M)=5.6, FLOPs (G)=2.03, Resolution=256x2562023.03 | 78.4 | 1,039 | 56.7 | 941 | |
| PoolFormer-S12Params (M)=11.9, FLOPs (G)=1.82, Resolution=224x2242023.03 | 77.2 | 1,439 | 49 | 665 | |
| FasterNet-T1Params (M)=7.6, FLOPs (G)=0.85, Resolution=224x2242023.03 | 76.2 | 3,782 | 17.7 | 285 | |
| GhostNet x 1.3Params (M)=7.4, FLOPs (G)=0.24, Resolution=224x2242023.03 | 75.7 | 2,988 | 17.9 | 481 | |
| PVT-TinyParams (M)=13.2, FLOPs (G)=1.94, Resolution=224x2242023.03 | 75.1 | 1,266 | 55.6 | 708 | |
| EdgeNeXt-XSParams (M)=2.3, FLOPs (G)=0.54, Resolution=256x2562023.03 | 75 | 1,738 | 24.4 | 434 | |
| ShuffleNetV2 x 2Params (M)=7.4, FLOPs (G)=0.59, Resolution=224x2242023.03 | 74.9 | 3,339 | 17.8 | 403 | |
| MobileViT-XSParams (M)=2.3, FLOPs (G)=1.05, Resolution=256x2562023.03 | 74.8 | 1,392 | 40.8 | 648 | |
| MobileNetV2 x 1.4Params (M)=6.1, FLOPs (G)=0.6, Resolution=224x2242023.03 | 74.7 | 2,711 | 22.6 | 650 | |
| SHiNeBackbone=ViT-L/14, Hierarchy Source=LLM, Zero-shot=true, Batch size=1, GPU=NVIDIA RTX 20702024.05 | 73.6 | 81 | — | — | |
| CHiLSBackbone=ViT-L/14, Hierarchy Source=LLM, Zero-shot=true, Batch size=1, GPU=NVIDIA RTX 20702024.05 | 73.4 | 23 | — | — | |
| SHiNeBackbone=ViT-L/14, Hierarchy Source=WordNet (WDN), Zero-shot=true, Batch size=1, GPU=NVIDIA RTX 20702024.05 | 73.1 | 81 | — | — | |
| ShuffleNetV2 x 1.5Params (M)=3.5, FLOPs (G)=0.3, Resolution=224x2242023.03 | 72.6 | 4,898 | 12.1 | 266 | |
| CHiLSBackbone=ViT-L/14, Hierarchy Source=WordNet (WDN), Zero-shot=true, Batch size=1, GPU=NVIDIA RTX 20702024.05 | 72.1 | 23 | — | — | |
| MobileNetV2Params (M)=3.5, FLOPs (G)=0.31, Resolution=224x2242023.03 | 72 | 4,198 | 12.2 | 442 | |
| CLIPBackbone=ViT-L/14, Hierarchy Source=Vanilla, Zero-shot=true, Batch size=1, GPU=NVIDIA RTX 20702024.05 | 72 | 81 | — | — | |
| FasterNet-T0Params (M)=3.9, FLOPs (G)=0.34, Resolution=224x2242023.03 | 71.9 | 6,807 | 9.2 | 143 | |
| EdgeNeXt-XXSParams (M)=1.3, FLOPs (G)=0.26, Resolution=256x2562023.03 | 71.2 | 2,765 | 15.7 | 239 | |
| H-CLIPBackbone=ViT-L/14, Hierarchy Source=WordNet (WDN), Reproduction=Without uncertainty estimation and refined WordNet hierarchy, Zero-shot=true, Batch size=1, GPU=NVIDIA RTX 20702024.05 | 70.6 | 2 | — | — | |
| MobileViT-XXSParams (M)=1.3, FLOPs (G)=0.42, Resolution=256x2562023.03 | 69 | 2,393 | 30.8 | 348 | |
| H-CLIPBackbone=ViT-L/14, Hierarchy Source=LLM, Zero-shot=true, Batch size=1, GPU=NVIDIA RTX 20702024.05 | 66.9 | 1 | — | — | |
| SHiNeBackbone=ViT-B/16, Hierarchy Source=LLM, Zero-shot=true, Batch size=1, GPU=NVIDIA RTX 20702024.05 | 66.7 | 149 | — | — | |
| CHiLSBackbone=ViT-B/16, Hierarchy Source=LLM, Zero-shot=true, Batch size=1, GPU=NVIDIA RTX 20702024.05 | 66.1 | 27 | — | — | |
| SHiNeBackbone=ViT-B/16, Hierarchy Source=WordNet (WDN), Zero-shot=true, Batch size=1, GPU=NVIDIA RTX 20702024.05 | 65.5 | 150 | — | — | |
| CHiLSBackbone=ViT-B/16, Hierarchy Source=WordNet (WDN), Zero-shot=true, Batch size=1, GPU=NVIDIA RTX 20702024.05 | 64.6 | 28 | — | — | |
| CLIPBackbone=ViT-B/16, Hierarchy Source=Vanilla, Zero-shot=true, Batch size=1, GPU=NVIDIA RTX 20702024.05 | 63.9 | 152 | — | — | |
| H-CLIPBackbone=ViT-B/16, Hierarchy Source=WordNet (WDN), Reproduction=Without uncertainty estimation and refined WordNet hierarchy, Zero-shot=true, Batch size=1, GPU=NVIDIA RTX 20702024.05 | 63.8 | 3 | — | — | |
| SHiNeBackbone=ViT-B/32, Hierarchy Source=LLM, Zero-shot=true, Batch size=1, GPU=NVIDIA RTX 20702024.05 | 61.6 | 141 | — | — | |
| CHiLSBackbone=ViT-B/32, Hierarchy Source=LLM, Zero-shot=true, Batch size=1, GPU=NVIDIA RTX 20702024.05 | 61.1 | 26 | — | — | |
| SHiNeBackbone=ViT-B/32, Hierarchy Source=WordNet (WDN), Zero-shot=true, Batch size=1, GPU=NVIDIA RTX 20702024.05 | 60.3 | 142 | — | — | |
| H-CLIPBackbone=ViT-B/16, Hierarchy Source=LLM, Zero-shot=true, Batch size=1, GPU=NVIDIA RTX 20702024.05 | 60.1 | 2 | — | — | |
| CHiLSBackbone=ViT-B/32, Hierarchy Source=WordNet (WDN), Zero-shot=true, Batch size=1, GPU=NVIDIA RTX 20702024.05 | 59.6 | 27 | — | — | |
| CLIPBackbone=ViT-B/32, Hierarchy Source=Vanilla, Zero-shot=true, Batch size=1, GPU=NVIDIA RTX 20702024.05 | 58.9 | 150 | — | — | |
| H-CLIPBackbone=ViT-B/32, Hierarchy Source=WordNet (WDN), Reproduction=Without uncertainty estimation and refined WordNet hierarchy, Zero-shot=true, Batch size=1, GPU=NVIDIA RTX 20702024.05 | 58.7 | 3 | — | — | |
| H-CLIPBackbone=ViT-B/32, Hierarchy Source=LLM, Zero-shot=true, Batch size=1, GPU=NVIDIA RTX 20702024.05 | 55.8 | 2 | — | — |