Query-by-Example Retrieval on AudioSet (evaluation)
79mAPExplicit Label Triplet (topline)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Explicit Label Triplet (topline)Backbone=ResNet-50, Embedding Dimension=1282017.11 | 79 | 100 | |
| Joint Unsupervised TripletBackbone=ResNet-50, Embedding Dimension=1282017.11 | 57.5 | 41 | |
| Temporal Proximity (Δt=10s)Backbone=ResNet-50, Embedding Dimension=1282017.11 | 56.2 | 38 | |
| T/F Translation (S = 10)Backbone=ResNet-50, Embedding Dimension=1282017.11 | 50.8 | 23 | |
| Mixed Example (α=0.25)Backbone=ResNet-50, Embedding Dimension=1282017.11 | 48.9 | 18 | |
| Gaussian Noise (σ=0.5)Backbone=ResNet-50, Embedding Dimension=1282017.11 | 47.8 | 15 | |
| Log Mel Spectrogram (baseline)Embedding Dimension=61442017.11 | 42.3 | 0 |