Just detect "target" for object detection

[MinhxNguyen7]

Description

Since we're detecting objects to narrow down on targets, then classifying them, should we just train the first object detection model on where there are targets in general? Are we trying to eliminate letter and shape classification?

A middle ground could be to train only on shape detection since the shapes might be large enough to detect in one shot, and we have sufficient data for shapes, see this issue. (I just realized that this is the current method.) Then, we can perhaps combine color and letter classification into one YOLO classify model.

@EricPedley ?