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

This report describes our submission to the ActivityNet Challenge at CVPR2019. We use a 3D convolutional neural network (CNN) based front-end and anensemble of temporal convolution and LSTM classifiers to predict whether avisible person is speaking or not. Our results show significant improvementsover the baseline on the AVA-ActiveSpeaker dataset.