Fig. 10
From: Recalling of multiple grasping methods from an object image with a convolutional neural network
Results of recalling the grasping method with certainty for the training data. Each row in the second to rightmost columns corresponds to each typical grasping type that is clustered in the grasping position network. The third column displays the estimated grasping position and the ground truth in red and green, and the intersection pixels in yellow. The ground truth is displayed only on the image of the channel that is selected in Eq. (2). The pixel color of the object and the hand shape images encodes the depth values by the “red(near)-blur(far)” colormap