view article
Figure 1
Deep Consensus workflow. Deep Consensus takes the coordinates proposed by different particle pickers as input, from which the intersection (AND set) and the union (OR set) of these coordinates are computed. Next, it picks random coordinates providing that they do not overlap with the OR set (NEG set). The NEG and AND sets are then used to train a convolutional neural network (CNN) that will finally classify the coordinates of the OR set (which is the largest set) as positive particle coordinates or negative particle coordinates.

IUCrJ
Volume 5| Part 6| November 2018| Pages 854-865
ISSN: 2052-2525
https://doi.org/10.1107/S2052252518014392