LEARNING SOUND EVENTS FROM WEBLY LABELED DATA. PDF

Authors: Anurag Kumar, Ankit Shah, Alex Hauptmann, Bhiksha Raj

Home

Classwise Results

Webly-4k

N1-Self

N1-Self
Fig 1. - AP for sound events on Webly-4k training for Baseline (N1-Self) system. Mean Average Precision (MAP) also shown.

N2-Self

N2-Self
Fig 2. - AP for sound events on Webly-4k training for N2-Self. Mean Average Precision (MAP) also shown. Transfer learning helps improve the performance.

WeblyNet

baseline vs weblynet
Fig 3. - AP for sound events on Webly-4k training for WeblyNet. Mean Average Precision (MAP) also shown. Training two networks together helps in better learning.

Comparison of Baseline and WeblyNet

baseline vs weblynet
Fig 4. - AP for sound events on Webly-4k training. Comparison of baseline vs WeblyNet system. Mean Average Precision (MAP) also shown.

Training on Manually Labeled Data (Audioset-40)

baseline vs weblynet
Fig 4. - AP for sound events on Audioset-40 training. Performance of N1-Self. Mean Average Precision (MAP) also shown.

Webly-2k

N1-Self

N1-Self
Fig 1. - AP for sound events on Webly-2k training for Baseline (N1-Self) system. Mean Average Precision (MAP) also shown.

N2-Self

N2-Self
Fig 2. - AP for sound events on Webly-2k training for N2-Self. Mean Average Precision (MAP) also shown. Transfer learning helps improve the performance.

WeblyNet

baseline vs weblynet
Fig 3. - AP for sound events on Webly-2k training for WeblyNet. Mean Average Precision (MAP) also shown. Training two networks together helps in better learning.