I have 3 types of images:
A: Images of apples
B: Images of bananas, however some of these banana images also contain apples in the observable background
Is it enough to just label the bounding boxes of the individual items or MUST I label images with both in? Rephrasing the above question, do I have to label all label-able objects in a scene, or can I leave some items out? How will this effect accuracy?
Will my model (YOLO) still be able to detect multiple objects in scene, having only been trained on images with one class label in scene?