How similar the outputs from VideoDetector vs FrameDetector supposed to be?
If we are charting the VideoDetector at 25fps, it produces clean lines, if charting the FrameDetector for the same content at same fps, we get huge fluctuations between two frames. 1/25sec frames following each other could differ 50-70 points which seems unreasonable.
We are evaluating the FrameDetector, as the VideoDetector was keep losing frames. Specified parameters for it to process the content at 25fps, but it was skipping/losing frames, resulting anywhere from 8 to 25 samples per second. We have switched to FrameDetector to be able to aggregate results across videos and getting the above issue now.