Real time issue #1
Description
One question. If your input is from a video and captured frame by frame, how long could this model be real time? I mean, if you keep importing photos into mask rcnn model and running the detect function, the graph will exceed device memory very soon. I tried wait and collect amount of frames as a batch, then do the detect and clear graph after each batch, but in this way, it will not be real time. Any solutions?