Image-Style-Transfer-using-AlexNet

This project proposes an improved AlexNet image style transfer model, which has larger convolution core to capture more details of image features, and more complex network parameters compared with the traditional VGG network, aiming at solving the problems of long iterative optimization time and poor image quality in previous studies. On this foundation, the exponential linear unit ELU combined with the activation value offset of Gram matrix is introduced into the AlexNet network, suppressing the artifact phenomenon to a certain degree compared with the original AlexNet image style transfer model. What’s more, compared with the other four main image style transfer models (Realized in other four ipynb files), the method in this project takes an average of about 2 minutes, the image contrast is 793.41, the SSIM and PSNR are 0.3804 and 13.496 respectively, which has been better lifted in processing speed, picture definition and vision art effect. Finally, this project integrates the improved AlexNet image style transfer model into the form of an interactive PC applet through Python Tkinter package and pyinstaller method. Users can conduct style transfer by selecting arbitrary pictures and specifying pixels by themselves