Fig. 2
From: TasselNetv2: in-field counting of wheat spikes with context-augmented local regression networks
Three examples of incomplete objects when only looking at the local patches. White parts are invisible contextual regions for the current visible patches. Wheat spikes annotated with black dots indicate the spike is partly within the visible area, and red dots represent spikes with severe occlusions. In both cases, accurate wheat numbers are just hard to obtain without the help of local visual context