Despite the progress seen in classification methods, current approaches for
handling videos with distribution shifts in source and target domains remain
source-dependent as they require access to the source data during the
adaptation stage. In this paper, we present a self-training bas