notesum.ai
Published at November 15Unlocking Transfer Learning for Open-World Few-Shot Recognition
cs.CV
cs.AI
Released Date: November 15, 2024
Authors: Byeonggeun Kim1, Juntae Lee1, Kyuhong Shim1, Simyung Chang1
Aff.: 1Qualcomm AI Research

| Methods | 1-shot | 5-shot | |||
| Acc. | AUROC | Acc. | AUROC | ||
| miniImageNet | |||||
| ProtoNets[37]† | 64.01 0.9 | 51.81 0.9 | 80.09 0.6 | 60.39 0.9 | |
| FEAT [42]† | 67.02 0.9 | 57.01 0.8 | 82.02 0.5 | 63.18 0.8 | |
| PEELER [24]† | 65.86 0.9 | 60.57 0.8 | 80.61 0.6 | 67.35 0.8 | |
| SnaTCHer-F [16] | 67.02 0.9 | 68.27 1.0 | 82.02 0.5 | 77.42 0.7 | |
| SnaTCHer-T [16] | 66.60 0.8 | 70.17 0.9 | 81.77 0.5 | 76.66 0.8 | |
| SnaTCHer-L [16] | 67.60 0.8 | 69.40 0.9 | 82.36 0.6 | 76.15 0.8 | |
| ATT [15] | 67.64 0.8 | 71.35 0.7 | 82.31 0.5 | 79.85 0.6 | |
| ATT-G [15] | 68.11 0.8 | 72.41 0.7 | 83.12 0.5 | 79.85 0.6 | |
| GEL [41] | 68.26 0.9 | 73.70 0.8 | 83.05 0.6 | 82.29 0.6 | |
| OAL-OFL | 69.78 0.8 | 73.88 0.7 | 85.49 0.7 | 83.13 0.6 | |
| OAL-OFL-Lite | 69.15 0.8 | 72.21 0.9 | 85.61 0.6 | 81.11 0.6 | |
| tieredImageNet | |||||
| ProtoNets[37]† | 68.26 1.0 | 60.73 0.8 | 83.40 0.7 | 64.96 0.8 | |
| FEAT [42]† | 70.52 1.0 | 63.54 0.7 | 84.74 0.7 | 70.74 0.8 | |
| PEELER [24]† | 69.51 0.9 | 65.20 0.8 | 84.10 0.7 | 73.27 0.7 | |
| SnaTCHer-F [16] | 70.52 1.0 | 74.28 0.8 | 84.74 0.7 | 82.02 0.6 | |
| SnaTCHer-T [16] | 70.45 1.0 | 74.84 0.8 | 84.42 0.7 | 82.03 0.7 | |
| SnaTCHer-L [16] | 70.85 1.0 | 74.95 0.8 | 85.23 0.6 | 80.81 0.7 | |
| ATT [15] | 69.34 1.0 | 72.74 0.8 | 83.82 0.6 | 78.66 0.7 | |
| ATT-G [15] | 70.58 0.9 | 73.43 0.8 | 85.38 0.6 | 81.64 0.6 | |
| GEL [41] | 70.50 0.9 | 75.86 0.8 | 84.60 0.7 | 81.95 0.7 | |
| OAL-OFL | 71.73 0.5 | 75.88 0.6 | 86.75 0.6 | 83.36 0.6 | |
| OAL-OFL-Lite | 70.80 0.9 | 73.67 0.7 | 86.66 0.6 | 82.22 0.6 | |