Image Classification on iNaturalist 2018 (All)
74.6Top-1 AccuracyRIDE (6 experts)
Evaluation Results
| Method | Links | |
|---|---|---|
| RIDE (6 experts)GFlops=9.83 (2.4x), Training Epochs=300, Expert assignment module=None2020.10 | 74.6 | |
| iBOT-vMF (kp)Backbone=ViT-Base/16, Unique Protos.=7573 (92.4%), Evaluation Protocol=Fine-tuned2025.10 | 74 | |
| iBOT-vMF (kd)Backbone=ViT-Base/16, Unique Protos.=1634 (19.9%), Evaluation Protocol=Fine-tuned2025.10 | 73.3 | |
| RIDE (4 experts)GFlops=4.73 (1.2x), Training Epochs=2002020.10 | 73.2 | |
| RIDE (4 experts)GFlops=4.51 (1.1x)2020.10 | 72.6 | |
| RIDE (3 experts)GFlops=4.17 (1.0x)2020.10 | 72.2 | |
| CARP + DecouplingBackbone=ViT-Base/16, Unique Protos.=65536 (100%), Evaluation Protocol=Fine-tuned2025.10 | 71.8 | |
| CARP + DecouplingBackbone=ViT-Small/16, Unique Protos.=65536 (100%), Evaluation Protocol=Fine-tuned2025.10 | 71.7 | |
| RIDE (2 experts)GFlops=3.67 (0.9x)2020.10 | 71.4 | |
| iBOT-vMFBackbone=ViT-Small/16, Unique Protos.=1804 (22.0%), Evaluation Protocol=Fine-tuned2025.10 | 69.4 | |
| iBOT-vMF (kp)Backbone=ViT-Small/16, Unique Protos.=7895 (96.4%), Evaluation Protocol=Fine-tuned2025.10 | 69.3 | |
| iBOT-vMF (kd)Backbone=ViT-Small/16, Unique Protos.=1843 (22.5%), Evaluation Protocol=Fine-tuned2025.10 | 69.1 | |
| DINO-vMFBackbone=ViT-Small/16, Unique Protos.=1380 (2.1%), Evaluation Protocol=Fine-tuned2025.10 | 68.5 | |
| BBNGFlops=4.36 (1.1x)2020.10 | 66.3 | |
| LWSGFlops=4.14 (1.0x)2020.10 | 65.9 | |
| τ-normGFlops=4.14 (1.0x)2020.10 | 65.6 | |
| cRTGFlops=4.14 (1.0x)2020.10 | 65.2 | |
| LDAM + DRWGFlops=4.14 (1.0x)2020.10 | 64.6 | |
| PMSN (λ = 5)Backbone=ViT-Small/16, Unique Protos.=3005 (36.9%), Evaluation Protocol=Fine-tuned2025.10 | 64.2 | |
| OLTRGFlops=4.14 (1.0x)2020.10 | 63.9 | |
| MSN (λ = 1)Backbone=ViT-Small/16, Unique Protos.=3363 (41.3%), Evaluation Protocol=Fine-tuned2025.10 | 63.5 | |
| CEGFlops=4.14 (1.0x)2020.10 | 61.7 | |
| CB-FocalGFlops=4.14 (1.0x)2020.10 | 61.1 | |
| MSN (λ = 1)Backbone=ViT-Small/16, Unique Protos.=3363 (41.3%), Evaluation Protocol=Linear2025.10 | 53.8 | |
| iBOT-vMF (kp)Backbone=ViT-Base/16, Unique Protos.=7573 (92.4%), Evaluation Protocol=Linear2025.10 | 51.4 | |
| iBOT-vMF (kp)Backbone=ViT-Small/16, Unique Protos.=7895 (96.4%), Evaluation Protocol=Linear2025.10 | 51.1 | |
| iBOT-vMF (kd)Backbone=ViT-Small/16, Unique Protos.=1843 (22.5%), Evaluation Protocol=Linear2025.10 | 50.5 | |
| iBOT-vMF (kd)Backbone=ViT-Base/16, Unique Protos.=1634 (19.9%), Evaluation Protocol=Linear2025.10 | 50.4 | |
| iBOT-vMFBackbone=ViT-Small/16, Unique Protos.=1804 (22.0%), Evaluation Protocol=Linear2025.10 | 50.1 | |
| DINO-vMFBackbone=ViT-Small/16, Unique Protos.=1380 (2.1%), Evaluation Protocol=Linear2025.10 | 49.7 | |
| CARP + DecouplingBackbone=ViT-Small/16, Unique Protos.=65536 (100%), Evaluation Protocol=Linear2025.10 | 49.1 | |
| CARP + DecouplingBackbone=ViT-Base/16, Unique Protos.=65536 (100%), Evaluation Protocol=Linear2025.10 | 48.3 | |
| PMSN (λ = 5)Backbone=ViT-Small/16, Unique Protos.=3005 (36.9%), Evaluation Protocol=Linear2025.10 | 41.8 |