Image Classification on VTAB Natural Set
76.9AccuracyLiT ViT-e
Evaluation Results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| LiT ViT-eZero-shot=true2022.09 | 76.9 | — | — | — | — | — | — | — | — | |
| LiT ViT-ezero-shot=true, OCR-input=No2022.09 | 76.9 | 79.8 | 90.4 | 68.8 | 91.2 | 98.1 | 76.3 | 33.8 | — | |
| LiTTraining Data Source=Private, Evaluation Protocol=Zero-shot2021.11 | 74.7 | — | — | — | — | — | — | — | — | |
| LiT ViT-gZero-shot=true2022.09 | 74.7 | — | — | — | — | — | — | — | — | |
| LiT ViT-gzero-shot=true, OCR-input=No2022.09 | 74.7 | 79.2 | 83.6 | 66.6 | 92.3 | 97.7 | 76 | 27.5 | — | |
| CLIPZero-shot=true2022.09 | 73.9 | — | — | — | — | — | — | — | — | |
| CLIPzero-shot=true, OCR-input=Yes (for SVHN task)2022.09 | 73.9 | 92.8 | 77.5 | 55.7 | 78.3 | 93.5 | 68.4 | 51 | — | |
| ResNet50Training Data Source=Multiple datasets, Evaluation Protocol=Supervised fine-tuned2021.11 | 72.6 | — | — | — | — | — | — | — | — | |
| LiTTraining Data Source=Public, Evaluation Protocol=Zero-shot2021.11 | 63.1 | — | — | — | — | — | — | — | — | |
| CLIP-MoEBackbone=ViT-B/32, Evaluation=Zero-shot2026.04 | — | — | — | — | — | — | — | — | 58.8 | |
| Cluster-aware UpcyclingBackbone=ViT-B/32, Evaluation=Zero-shot2026.04 | — | — | — | — | — | — | — | — | 59.1 | |
| DeRS-LMBackbone=ViT-B/32, Evaluation=Zero-shot2026.04 | — | — | — | — | — | — | — | — | 58.1 | |
| Drop-SVDBackbone=ViT-B/32, Evaluation=Zero-shot2026.04 | — | — | — | — | — | — | — | — | 58.6 | |
| Drop-UpcyclingBackbone=ViT-B/32, r=0.75, Evaluation=Zero-shot2026.04 | — | — | — | — | — | — | — | — | 58.6 | |
| Drop-UpcyclingBackbone=ViT-B/32, r=0.50, Evaluation=Zero-shot2026.04 | — | — | — | — | — | — | — | — | 57.8 | |
| Drop-UpcyclingBackbone=ViT-B/32, r=0.25, Evaluation=Zero-shot2026.04 | — | — | — | — | — | — | — | — | 57.7 | |
| Sparse UpcyclingBackbone=ViT-B/32, Evaluation=Zero-shot2026.04 | — | — | — | — | — | — | — | — | 58 |