Fig. 1. VAE's basic framework

Fig. 2. Discriminator model

Fig. 3. Conditional generation adversarial networks based on VAE

Fig. 4. Total loss

Fig. 5. Sketches in multiple styles. (a) Result by Sobel operator; (b) result by taking the inverse and performing Gaussian blur; (c) result by Median filter and Laplace filter; (d) sketch in CUHK dataset

Fig. 6. Flow chart of model training process

Fig. 7. Synthesis results of face photos. (a) Test sketches; (b) images generated by the method in Ref. [4]; (c) images generated by the method in Ref. [7]; (d) images generated by the method in Ref. [9]; (e) images generated by our method; (f) real photos

Fig. 8. Generation results of face photos. (a) Test sketches; (b) generation results of test set by the single style generated network; (c) generation results of test set by universal style network model; (d) real photos