如何在使用更快的 rcnn/ssd 模型的同时加速对象检测

Question

I've got 50 videos spanning 90 minutes each. Now I'm running SSD for object detection and saving the frames with objects and their timestamps in a csv file. But the problem is that it still takes entire 90 minutes for each of the videos. My question :- Is there any way by which I can somehow kind of fast forward the SSD object detection model so that each video gets done in let's say 10/20 minutes maybe ?

Answer 1

Is every single frame really matter to you? If not, just run the detection every x frames only, not on every frame. for example use:

if (frame%2 == 0)
{
run detection
}

for taking only every two frames (even number frame)