Long Short-Term Transformer for Online Action Detection Supplementary Material

A Can we compensate reduced memory length with RNN? Cont'd To better understand the design choice in Sec. Figure 1a shows some qualitative results. The top sample contains the "Billiards"action that incurs only tiny motion. The bottom sample is challenging -- the "Open door" action occurs behind the female reporter and is Red circle indicates where the action is happening in each frame.