Image Classification on ImageNet Matched Frequency V2
80.8Top-1 AccFixEfficientNet-L2
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| FixEfficientNet-L2params=480M, train res=475, test res=600, extra-training data=300M additional unlabeled images, single crop=true2020.03 | 80.8 | 96.1 | — | |
| FixEfficientNet-L2size=480.3M, crop=single, external data=300M additional unlabeled images, training strategy=Fixing Resolution2020.03 | 80.8 | 96.1 | — | |
| EfficientNet-L2size=480.3M, crop=single, external data=300M additional unlabeled images, training strategy=Noisy Student2020.03 | 80.5 | 95.7 | — | |
| EfficientNet-L2params=480M, train res=475, test res=800, extra-training data=300M additional unlabeled images, single crop=true2020.03 | 80.3 | 95.8 | — | |
| FixEfficientNet-B7params=66M, train res=600, test res=632, extra-training data=300M additional unlabeled images, single crop=true2020.03 | 78.6 | 94.7 | — | |
| FixEfficientNet-B7size=66.4M, crop=single, external data=300M additional unlabeled images, training strategy=Fixing Resolution2020.03 | 78.6 | 94.7 | — | |
| EfficientNet-B7params=66M, train res=600, test res=600, extra-training data=300M additional unlabeled images, single crop=true2020.03 | 78.5 | 94.4 | — | |
| EfficientNet-B7size=66.4M, crop=single, external data=300M additional unlabeled images, training strategy=Noisy Student2020.03 | 78.5 | 94.4 | — | |
| FixResNeXt-101 32x48dsize=829.0M, crop=single, external data=Billion-scale, training strategy=Fixing Resolution2020.03 | 77.8 | 93.9 | — | |
| FixEfficientNet-B6params=43M, train res=528, test res=680, extra-training data=300M additional unlabeled images, single crop=true2020.03 | 77.5 | 94.3 | — | |
| FixEfficientNet-B6size=43.0M, crop=single, external data=300M additional unlabeled images, training strategy=Fixing Resolution2020.03 | 77.5 | 94.3 | — | |
| EfficientNet-B6params=43M, train res=528, test res=528, extra-training data=300M additional unlabeled images, single crop=true2020.03 | 77.3 | 93.9 | — | |
| EfficientNet-B6size=43.0M, crop=single, external data=300M additional unlabeled images, training strategy=Noisy Student2020.03 | 77.3 | 93.9 | — | |
| FixEfficientNet-B5params=30M, train res=456, test res=576, extra-training data=300M additional unlabeled images, single crop=true2020.03 | 77 | 94 | — | |
| FixEfficientNet-B5size=30.4M, crop=single, external data=300M additional unlabeled images, training strategy=Fixing Resolution2020.03 | 77 | 94 | — | |
| EfficientNet-B5params=30M, train res=456, test res=456, extra-training data=300M additional unlabeled images, single crop=true2020.03 | 76.8 | 93.6 | — | |
| EfficientNet-B5size=30.4M, crop=single, external data=300M additional unlabeled images, training strategy=Noisy Student2020.03 | 76.8 | 93.6 | — | |
| FixResNeXt-101 32x16dsize=193.0M, crop=single, external data=Billion-scale, training strategy=Fixing Resolution2020.03 | 76.7 | 93.4 | — | |
| FixResNeXt-101 32x8dsize=88.0M, crop=single, external data=Billion-scale, training strategy=Fixing Resolution2020.03 | 76.3 | 93.4 | — | |
| FixEfficientNet-B4params=19M, train res=380, test res=472, extra-training data=300M additional unlabeled images, single crop=true2020.03 | 76.2 | 93.6 | — | |
| FixEfficientNet-B4size=19.3M, crop=single, external data=300M additional unlabeled images, training strategy=Fixing Resolution2020.03 | 76.2 | 93.6 | — | |
| EfficientNet-B7Number of Parameters=66M, Training Resolution=600, Testing Resolution=6002020.03 | 76.1 | 93 | — | |
| EfficientNet-B8Number of Parameters=87.4M, Training Resolution=672, Testing Resolution=6722020.03 | 76.1 | 92.7 | — | |
| FixEfficientNet-B8Number of Parameters=87.4M, Training Resolution=672, Testing Resolution=8002020.03 | 75.9 | 93 | — | |
| FixEfficientNet-B7Number of Parameters=66M, Training Resolution=600, Testing Resolution=6322020.03 | 75.8 | 93.2 | — | |
| EfficientNet-B4params=19M, train res=380, test res=380, extra-training data=300M additional unlabeled images, single crop=true2020.03 | 75.7 | 93.1 | — | |
| EfficientNet-B4size=19.3M, crop=single, external data=300M additional unlabeled images, training strategy=Noisy Student2020.03 | 75.7 | 93.1 | — | |
| ResNeXt-101 32x16dsize=193.0M, crop=single, external data=Billion-scale, training strategy=Billion-scale2020.03 | 75.6 | 93.3 | — | |
| ResNeXt-101 32x8dsize=88.0M, crop=single, external data=Billion-scale, training strategy=Billion-scale2020.03 | 75.5 | 92.8 | — | |
| EfficientNet-B6Number of Parameters=43M, Training Resolution=528, Testing Resolution=5282020.03 | 75.4 | 92.4 | — | |
| FixEfficientNet-B6Number of Parameters=43M, Training Resolution=528, Testing Resolution=5762020.03 | 75.4 | 92.6 | — | |
| DeiT-B↑384⚗ / 1000 epochsimage size=384x384, distillation=true, epochs=10002020.12 | 75.2 | — | — | |
| FixEfficientNet-B5Number of Parameters=30M, Training Resolution=456, Testing Resolution=5762020.03 | 75.1 | 92.4 | — | |
| FixEfficientNet-B3params=12M, train res=300, test res=472, extra-training data=300M additional unlabeled images, single crop=true2020.03 | 75 | 93 | — | |
| FixEfficientNet-B3size=12.2M, crop=single, external data=300M additional unlabeled images, training strategy=Fixing Resolution2020.03 | 75 | 93 | — | |
| DeiT-B↑384⚗image size=384x384, distillation=true2020.12 | 74.8 | — | — | |
| FixResNeXt-101 32x4dsize=42.0M, crop=single, external data=Billion-scale, training strategy=Fixing Resolution2020.03 | 74.6 | 92.7 | — | |
| EfficientNet-B5Number of Parameters=30M, Training Resolution=456, Testing Resolution=4562020.03 | 74.6 | 92 | — | |
| ResNeXt-101 32x4dsize=42.0M, crop=single, external data=Billion-scale, training strategy=Billion-scale2020.03 | 74.2 | 92 | — | |
| EfficientNet-B3params=12M, train res=300, test res=300, extra-training data=300M additional unlabeled images, single crop=true2020.03 | 73.9 | 91.9 | — | |
| EfficientNet-B3size=12.2M, crop=single, external data=300M additional unlabeled images, training strategy=Noisy Student2020.03 | 73.9 | 91.9 | — | |
| FixEfficientNet-B4Number of Parameters=19M, Training Resolution=380, Testing Resolution=5122020.03 | 73.9 | 91.8 | — | |
| EfficientNet-B6image size=528x5282020.12 | 73.9 | — | — | |
| DeiT-B⚗ / 1000 epochsimage size=224x224, distillation=true, epochs=10002020.12 | 73.9 | — | — | |
| FixEfficientNet-B2params=9.2M, train res=260, test res=420, extra-training data=300M additional unlabeled images, single crop=true2020.03 | 73.6 | 92 | — | |
| FixEfficientNet-B2size=9.1M, crop=single, external data=300M additional unlabeled images, training strategy=Fixing Resolution2020.03 | 73.6 | 92 | — | |
| EfficientNet-B5image size=456x4562020.12 | 73.6 | — | — | |
| DeiT-B⚗image size=224x224, distillation=true2020.12 | 73.2 | — | — | |
| DATBackbone=ViT-B/16, Pre-training=ImageNet-21K, Fine-tuning=ImageNet-1K, Input Resolution=224x2242023.11 | 73.2 | — | — | |
| EfficientNet-B4Number of Parameters=19M, Training Resolution=380, Testing Resolution=3802020.03 | 72.9 | 91 | — | |
| FixEfficientNet-B1params=7.8M, train res=240, test res=384, extra-training data=300M additional unlabeled images, single crop=true2020.03 | 72.7 | 91.4 | — | |
| FixEfficientNet-B1size=7.8M, crop=single, external data=300M additional unlabeled images, training strategy=Fixing Resolution2020.03 | 72.7 | 91.4 | — | |
| FixEfficientNet-B3Number of Parameters=12M, Training Resolution=300, Testing Resolution=4722020.03 | 72.7 | 90.9 | — | |
| ResNeXt-50 32x4dsize=25.1M, crop=single, external data=Billion-scale, training strategy=Billion-scale2020.03 | 72.6 | 90.9 | — | |
| RegNetY-16GFimage size=224x2242020.12 | 72.4 | — | — | |
| DeiT-B↑384image size=384x3842020.12 | 72.4 | — | — | |
| EfficientNet-B2params=9.2M, train res=260, test res=260, extra-training data=300M additional unlabeled images, single crop=true2020.03 | 72.3 | 91.1 | — | |
| EfficientNet-B2size=9.1M, crop=single, external data=300M additional unlabeled images, training strategy=Noisy Student2020.03 | 72.3 | 91.1 | — | |
| EfficientNet-B4image size=380x3802020.12 | 72.3 | — | — | |
| ResNet50size=25.6M, crop=single, external data=Billion-scale, training strategy=Billion-scale2020.03 | 71.7 | 90.5 | — | |
| DeiT-S⚗ / 1000 epochsimage size=224x224, distillation=true, epochs=10002020.12 | 71.7 | — | — | |
| DeiT-Bimage size=224x2242020.12 | 71.5 | — | — | |
| EfficientNet-B1params=7.8M, train res=240, test res=240, extra-training data=300M additional unlabeled images, single crop=true2020.03 | 70.9 | 90.1 | — | |
| EfficientNet-B1size=7.8M, crop=single, external data=300M additional unlabeled images, training strategy=Noisy Student2020.03 | 70.9 | 90.1 | — | |
| EfficientNet-B3Number of Parameters=12M, Training Resolution=300, Testing Resolution=3002020.03 | 70.9 | 89.4 | — | |
| DADBackbone=ViT-B/16, Pre-training=ImageNet-21K, Fine-tuning=ImageNet-1K, Input Resolution=224x2242023.11 | 70.9 | — | — | |
| FixEfficientNet-B2Number of Parameters=9.2M, Training Resolution=260, Testing Resolution=4202020.03 | 70.8 | 90.2 | — | |
| RegNetY-8GFimage size=224x2242020.12 | 70.8 | — | — | |
| EfficientNet-B3image size=300x3002020.12 | 70.6 | — | — | |
| FixEfficientNet-B1Number of Parameters=7.8M, Training Resolution=240, Testing Resolution=3842020.03 | 70.1 | 89.6 | — | |
| DeiT-S⚗image size=224x224, distillation=true2020.12 | 70 | — | — | |
| FixEfficientNet-B0params=5.3M, train res=224, test res=320, extra-training data=300M additional unlabeled images, single crop=true2020.03 | 69.4 | 89.6 | — | |
| FixEfficientNet-B0size=5.3M, crop=single, external data=300M additional unlabeled images, training strategy=Fixing Resolution2020.03 | 69.4 | 89.6 | — | |
| RegNetY-4GFimage size=224x2242020.12 | 69.4 | — | — | |
| EfficientNet-B2Number of Parameters=9.2M, Training Resolution=260, Testing Resolution=2602020.03 | 68.9 | 88.4 | — | |
| EfficientNet-B2image size=260x2602020.12 | 68.8 | — | — | |
| DeiT-Simage size=224x2242020.12 | 68.5 | — | — | |
| FixEfficientNet-B0Number of Parameters=5.3M, Training Resolution=224, Testing Resolution=3202020.03 | 67.8 | 87.9 | — | |
| EfficientNet-B0params=5.3M, train res=224, test res=224, extra-training data=300M additional unlabeled images, single crop=true2020.03 | 67.7 | 88.1 | — | |
| EfficientNet-B0size=5.3M, crop=single, external data=300M additional unlabeled images, training strategy=Noisy Student2020.03 | 67.7 | 88.1 | — | |
| EfficientNet-B1Number of Parameters=7.8M, Training Resolution=240, Testing Resolution=2402020.03 | 67.5 | 87.8 | — | |
| ResNet-152image size=224x2242020.12 | 67 | — | — | |
| EfficientNet-B1image size=240x2402020.12 | 66.9 | — | — | |
| ResNet-101image size=224x2242020.12 | 65.7 | — | — | |
| ViTBackbone=ViT-B/16, Pre-training=ImageNet-21K, Fine-tuning=ImageNet-1K, Input Resolution=224x2242023.11 | 65.7 | — | — | |
| EfficientNet-B0Number of Parameters=5.3M, Training Resolution=224, Testing Resolution=2242020.03 | 65.5 | 85.6 | — | |
| DeiT-Ti⚗ / 1000 epochsimage size=224x224, distillation=true, epochs=10002020.12 | 65.4 | — | — | |
| EfficientNet-B0image size=224x2242020.12 | 64.3 | — | — | |
| ResNet-50image size=224x2242020.12 | 63.3 | — | — | |
| DeiT-Ti⚗image size=224x224, distillation=true2020.12 | 62.9 | — | — | |
| DeiT-Tiimage size=224x2242020.12 | 60.4 | — | — | |
| ResNet-18image size=224x2242020.12 | 57.1 | — | — | |
| DINORANKCLIP (full)Evaluation Protocol=Zero-shot2026.05 | 10.4 | — | — | |
| Ours w/o DINOEvaluation Protocol=Zero-shot2026.05 | 9.65 | — | — | |
| RANKCLIPEvaluation Protocol=Zero-shot2026.05 | 9.01 | — | — | |
| ALIPEvaluation Protocol=Zero-shot2026.05 | 7.82 | — | — | |
| CyCLIPEvaluation Protocol=Zero-shot2026.05 | 7.68 | — | — | |
| CLIPEvaluation Protocol=Zero-shot2026.05 | 7.53 | — | — | |
| ACModel Ensemble=Linear models over CLIP representations2024.04 | — | — | 15.53 | |
| ALine-DModel Ensemble=Linear models over CLIP representations2024.04 | — | — | 8.44 |