Audio Classification on ESC-50 (val)
99Top-1 AccAST
Evaluation Results
| Method | Links | |
|---|---|---|
| ASTFeatures=Mel Spectrogram, Training Setup=Audioset + ESC, Last.L2026.02 | 99 | |
| CNN-1Features=MFCC + GTCC + CH + LM, Training Setup=ESC + US8K, Last.L2026.02 | 92 | |
| CNN-2Features=MFCC + GTCC + CH + LM, Training Setup=ESC + US8K, Last.L2026.02 | 87 | |
| VICRegBackbone=ResNet-18, Evaluation Protocol=Linear classifier on frozen representations2021.05 | 78.4 | |
| Barlow TwinsBackbone=ResNet-18, Evaluation Protocol=Linear classifier on frozen representations2021.05 | 75.4 | |
| Supervised baselineBackbone=ResNet-18, Evaluation Protocol=Supervised training on time-frequency representation2021.05 | 72.7 | |
| CNN-1Features=LM, Training Setup=ESC, All.L2026.02 | 68 | |
| CNN-2Features=LM + TZ, Training Setup=ESC, All.L2026.02 | 66 | |
| ASTFeatures=Mel Spectrogram, Training Setup=ESC + US8K, Last.L2026.02 | 58 | |
| ASTFeatures=Mel Spectrogram, Training Setup=ESC, All.L2026.02 | 43 |