Closed Jacobew closed 3 years ago
Currently, we do not have a plan for Youtube-8M. For Youtube-8M classification, algorithms should focus on how to aggregate the frame-level feature. You can refer to works like NetVLAD for Youtube-8M classification. Besides, training an MLP or linear-SVM with video-level features also serves as a baseline.
Hi, do you have any plan of supporting training for Youtube-8M? This is a multi-label video classification dataset of large scale, and it releases two kinds of features, i.e., frame-level (1.5T) and video-level features (30+G). How can I train on this dataset with the already available features?