Supplementary Material for " Dynamic Normalization and Relay for Video Action Recognition "

We evaluate our DNR method on four action recognition datasets following standard training/testing splits and protocols. Kinetics-400 [1] is a large scale action recognition dataset with trimmed video clips of around 10-second durations. It is collected from realistic YouTube videos, which covers 400 categories of human activities. In total, it contains around 240K training videos and 20K validation videos. Because of the expirations of some YouTube video links, the sizes of kinetics-400 dataset copies used in existing works may be different.