Fig. 3From: A novel comparative deep learning framework for facial age estimationOptimization of our CRCNN approach: performance by different setting of the deep architecture’s parameters. a Fusion strategy. b Baseline. c Region detection. d CONV. layers. e LOCAL layers. f FULL layers. g Batch size. h Activation function. i Dropout. j Learning rate. k Momentum. l Weight PenaltyBack to article page