There are two things I would like to see in VTM that would improve the automatic thumbnail selection process.
1. Sharpness metric. Sometimes VTM selects a frame that is part of a motion or a transition. These frames are often blurry and should therefore not be selected. I think it is technically possible to determine the sharpness of a frame and choosing a frame that is not blurry.
2. Face detection. Usually, a frame with a face in the right size and from the right direction is a good candidate for a thumbnail. Since there are many libraries for face detection available (like OpenCV) it should be possible to implement such a technique. This should be opt-in.
Even for manually thumbnail selection, both suggestions could improve user experience.
