Age Estimation Method based on Comparative Convolutional Neural Network using Inception Module

Enhanced Cycle Generative Adversarial Network for Generating Face Images of Untrained Races and Ages for Age Estimation

“…In label-sensitive deep metric learning (LSDML) [12] method, two paired images (same age class) and two unpaired images (different age classes) were selected to construct a mini-batch of data, with a total of four images. Another method for age estimation is a comparative CNN (CCNNAE) [14]. In this study, Inception-v3 [15] was used for the backbone model, and 70-dimensional feature vectors of the input images were extracted in the last layer.…”

Enhancing the Accuracies of Age Estimation With Heterogeneous Databases Using Modified CycleGAN

“…Fake A is an image output from the second generator in A → B → A, while Fake B is an image output from the second generator in B → A → B. The generator loss of the modified CycleGAN is expressed by including all the generator losses of B → A → B shown in Equations (2)- (5). B → A → B should be calculated in the same way as Equations (2)- (5).…”

“…The generator loss of the modified CycleGAN is expressed by including all the generator losses of B → A → B shown in Equations (2)- (5). B → A → B should be calculated in the same way as Equations (2)- (5). Equation (6) shows the final generator loss including all the previous losses.…”

“…-The generator of original CycleGAN adopts L G (Equation 6 C. CCNNAE Figure 7 shows the structure of CCNNAE [5] used in this study. The basic model of CCNNAE is Inception-v3 [37].…”

“…The basic model of CCNNAE is Inception-v3 [37]. For existing age estimation methods, the loss function commonly uses the Softmax cross entropy loss, while the similarity of the two images is measured in mini-batch to form the loss function in CCNNAE [5]. The basic Inception-v3 consists of only one model.…”