Alert button

Naver at ActivityNet Challenge 2019 -- Task B Active Speaker Detection (AVA)

Add code
Bookmark button
Alert button
Jun 25, 2019
Joon Son Chung

Figure 1 for Naver at ActivityNet Challenge 2019 -- Task B Active Speaker Detection (AVA)
Figure 2 for Naver at ActivityNet Challenge 2019 -- Task B Active Speaker Detection (AVA)
Figure 3 for Naver at ActivityNet Challenge 2019 -- Task B Active Speaker Detection (AVA)
Figure 4 for Naver at ActivityNet Challenge 2019 -- Task B Active Speaker Detection (AVA)

Share this with someone who'll enjoy it:

Twitter IconFacebook IconLinkedin IconWhatsapp IconMessenger Icon

This report describes our submission to the ActivityNet Challenge at CVPR 2019. We use a 3D convolutional neural network (CNN) based front-end and an ensemble of temporal convolution and LSTM classifiers to predict whether a visible person is speaking or not. Our results show significant improvements over the baseline on the AVA-ActiveSpeaker dataset.

View paper onarxiv icon

Share this with someone who'll enjoy it:

Twitter IconFacebook IconLinkedin IconWhatsapp IconMessenger Icon