I got an image processing project which gets an input video that could be lengthy and returns the frames which camera shots change in the input video. So we can have a summary of shots in whole video.
I have attached two images that belong to one shot. as you can see the only difference between two images is the flashlight of some camera during the shot. So as you might guess, machine observes this flashlight and thinks that shot has changed.
how can I teach my CNN model to ignore this immediate change in same shot?