Image Classification on iNaturalist 2019 (test)
86.577Top-1 AccuracyFixRes Ensemble (iNaturalist)
Evaluation results
| Method | Links | ||
|---|---|---|---|
| FixRes Ensemble (iNaturalist)Models=SE-ResNext-101-32x4d, SENet-154, Inception-ResNet-V2, ResNet-152-MPN-COV [22], Training Resolutions=448, 448, 491, 448, Test Resolutions=704, 672, 681, 448, Fine-tuning Layer 4=SE-ResNext-101-32x4d, SENet-154, Fine-tuning Classifier=All models, Fine-tuning Batch-norm=All models2019.06 | 86.577 | — | |
| ViT-B/16Pretraining + MaskSub=true, Finetuning + MaskSub=true2023.06 | 79.1 | — | |
| ViT-B/16Pretraining + MaskSub=true, Finetuning + MaskSub=false2023.06 | 78.8 | — | |
| ViT-B/16Pretraining + MaskSub=false, Finetuning + MaskSub=false2023.06 | 78.5 | — | |
| DeiT-BParams (M)=86.6, Fine-tuning=true2021.06 | 77.7 | — | |
| ViT-S/16Pretraining + MaskSub=true, Finetuning + MaskSub=true2023.06 | 77.1 | — | |
| ViT-S/16Pretraining + MaskSub=true, Finetuning + MaskSub=false2023.06 | 77 | — | |
| ViT-S/16Pretraining + MaskSub=false, Finetuning + MaskSub=false2023.06 | 76.7 | — | |
| ViTAE-SParams (M)=23.6, Fine-tuning=true2021.06 | 76 | — | |
| Grafit ResNet-50Params (M)=25.6, Fine-tuning=true2021.06 | 75.9 | — | |
| A2Backbone=ResNet-502021.10 | 75 | — | |
| DeiT-SParam(M)=22.05, GFLOPS=4.62023.04 | 74.8 | — | |
| dTPSBackbone=DeiT-S, Param(M)=22.77, GFLOPS=3.0, training epochs=1002023.04 | 74.7 | — | |
| dTPSBackbone=DeiT-S, Param(M)=22.77, GFLOPS=3.0, training epochs=302023.04 | 74.2 | — | |
| DynamicViTBackbone=DeiT-S, Param(M)=22.77, GFLOPS=2.9, training epochs=30, implementation=re-impl2023.04 | 74 | — | |
| A1Backbone=ResNet-502021.10 | 73.9 | — | |
| A3Backbone=ResNet-502021.10 | 73.8 | — | |
| ViTAE-TParams (M)=4.8, Fine-tuning=true2021.06 | 73.3 | — | |
| PytorchBackbone=ResNet-50, Pre-training recipe=[1]2021.10 | 73.2 | — | |
| DeiT-TParam(M)=5.72, GFLOPS=1.262023.04 | 72.8 | — | |
| dTPSBackbone=DeiT-T, Param(M)=5.90, GFLOPS=0.8, training epochs=1002023.04 | 72.4 | — | |
| dTPSBackbone=DeiT-T, Param(M)=5.90, GFLOPS=0.8, training epochs=302023.04 | 71.7 | — | |
| DynamicViTBackbone=DeiT-T, Param(M)=5.90, GFLOPS=0.8, training epochs=30, implementation=re-impl2023.04 | 71.4 | — |