Image Classification on VTAB (test)
65.4Avg AccuracyBig Transfer
Evaluation results
| Method | Links | ||||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Big TransferProtocol=fine-tuning, Net=RN50, Img=224, Param.=23.5M2022.06 | 65.4 | 69.4 | 81 | 54.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BiTBackbone=RN-18, Params (M)=11.2M, Image Size=2242021.07 | 64.3 | 68.5 | 80.8 | 53 | 84.6 | 47.1 | 82.7 | 83.9 | 29.1 | 83.4 | 93.8 | 74.1 | 80.7 | 74.5 | 55.2 | 58.7 | 98.6 | 46.5 | 20.1 | 21.8 | 43.7 | 78.8 | |
| UpperCaSEProtocol=hybrid, Net=ENB0, Img=224, Param.=4.0M2022.06 | 58.4 | 69.1 | 80.3 | 39.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| UpperCaSEProtocol=hybrid, Net=RN50, Img=224, Param.=23.5M2022.06 | 56.6 | 66.3 | 80.1 | 37.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MD-TransferProtocol=fine-tuning, Net=RN18, Img=126, Param.=11.2M2022.06 | 55.6 | 52.4 | 72.9 | 49.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MD-TransferBackbone=RN-18, Params (M)=11.2M, Image Size=1262021.07 | 55.6 | 52.4 | 72.9 | 49.4 | 70.6 | 31.3 | 66.1 | 49.1 | 13.9 | 83.2 | 88.7 | 63.7 | 81.5 | 57.6 | 40.3 | 52.9 | 85.9 | 46.4 | 36.5 | 31.2 | 43 | 58.7 | |
| Simple CNAPs + LITEBackbone=EN-B0, Params (M)=4.0M, Image Size=2242021.07 | 51.4 | 65.2 | 71.9 | 30.8 | 87.7 | 48.8 | 83.5 | 89.3 | 30.9 | 51 | 89.3 | 76.4 | 81.4 | 40.3 | 31.4 | 32.8 | 12.3 | 31.1 | 14.5 | 21 | 39.4 | 63.9 | |
| CTXBackbone=RN-34, Params (M)=21.3M, Image Size=2242021.07 | 50.5 | 61.1 | 67.3 | 34.1 | 84.2 | 37.5 | 81.8 | 70.9 | 24.8 | 67.2 | 86.4 | 67.7 | 79.8 | 35.5 | 27.9 | 29.6 | 23.2 | 46.9 | 37 | 21.6 | 31.9 | 54.3 | |
| ProtoMAMLBackbone=RN-18, Params (M)=11.2M, Image Size=1262021.07 | 45 | 45.7 | 70.7 | 31.5 | 73.1 | 29.7 | 60.2 | 56.6 | 8.1 | 46.8 | 80.1 | 53.5 | 75.9 | 73.2 | 32.7 | 35.4 | 42 | 23 | 13.4 | 18.8 | 32.5 | 54.4 | |
| Simple CNAPsBackbone=EN-B0, Params (M)=4.0M, Image Size=842021.07 | 44.7 | 49.5 | 63.8 | 31.7 | 79.6 | 37.1 | 65.5 | 69.8 | 18 | 26.7 | 82.8 | 64.5 | 78.4 | 29.4 | 30.7 | 32.5 | 43.9 | 21.1 | 13.5 | 19.6 | 33.9 | 58.1 | |
| SURProtocol=fine-tuning, Net=RN50, Img=224, Param.=164.6M2022.06 | 43.7 | 50.9 | 66.2 | 27.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SURBackbone=RN-50 x 7, Params (M)=164.6M, Image Size=224, Linear classifier head=true2021.07 | 43.7 | 50.9 | 66.2 | 27.2 | 86.5 | 34.2 | 71.2 | 88.7 | 0.5 | 24.2 | 82.6 | 67.8 | 77.1 | 37.4 | 34.1 | 29.8 | 16.9 | 18.7 | 8.3 | 18.4 | 33.5 | 57.5 | |
| ProtoNetsBackbone=RN-18, Params (M)=11.2M, Image Size=1262021.07 | 40.2 | 44 | 57.6 | 28.7 | 72 | 27.7 | 57.1 | 51 | 14.2 | 41.9 | 77.7 | 50.8 | 73.8 | 28 | 32 | 39.4 | 38.1 | 16.3 | 12.3 | 17.4 | 31.8 | 42.1 |