5-way 5-shot Classification on TieredImageNet (Outlier Noise Robustness)
71.42Accuracy (0% Noise)Oracle
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| OracleProtocol=5-way 5-shot2022.04 | 71.42 | 69.19 | 66.14 | 60.39 | |
| Vanilla ProtoNetBackbone=4-layer convolutional2022.04 | 71.42 | 67.58 | 60.97 | 50.29 | |
| Baseline++Backbone=4-layer convolutional2022.04 | 71.29 | 67.07 | 60.64 | 50.07 | |
| RNNPBackbone=4-layer convolutional2022.04 | 71.28 | 67.29 | 60.83 | 50.09 | |
| Spatial median prototypes (Median)Aggregation method=Median, Backbone=4-layer convolutional2022.04 | 71.28 | 67.79 | 61.63 | 50.63 | |
| Similarity weighted prototypes (Euclidean)Aggregation method=Euclidean distance weighting, Backbone=4-layer convolutional2022.04 | 71.28 | 67.89 | 61.61 | 50.49 | |
| Similarity weighted prototypes (Absolute)Aggregation method=Absolute distance weighting, Backbone=4-layer convolutional2022.04 | 71.17 | 68 | 61.98 | 50.59 | |
| TraNFS-3Number of Layers=3, Backbone=4-layer convolutional2022.04 | 71.13 | 67.93 | 62.39 | 51.82 | |
| TraNFS-2Number of Layers=2, Backbone=4-layer convolutional2022.04 | 70.83 | 67.52 | 61.76 | 51.4 | |
| Similarity weighted prototypes (Cosine)Aggregation method=Cosine similarity weighting, Backbone=4-layer convolutional2022.04 | 70.79 | 67.94 | 62.37 | 51.12 | |
| Linear ClassifierBackbone=4-layer convolutional2022.04 | 69.6 | 64.58 | 57.57 | 47.9 | |
| Matching NetworksBackbone=4-layer convolutional2022.04 | 64.99 | 60.74 | 54.28 | 44.93 | |
| MAMLBackbone=4-layer convolutional2022.04 | 63.9 | 58.14 | 51.89 | 42.01 | |
| Nearest k = 5k=5, Backbone=4-layer convolutional2022.04 | 59.25 | 55.3 | 49.34 | 40.56 | |
| Nearest k = 1k=1, Backbone=4-layer convolutional2022.04 | 58.89 | 54.57 | 49.45 | 43.2 | |
| Nearest k = 3k=3, Backbone=4-layer convolutional2022.04 | 58.38 | 53.98 | 48.06 | 40.11 |