Using a computer to generate images with realistic images is a new direction in current computer vision research. This paper designs an image generation model based on the Generative Adversarial Network (GAN). This paper creates a model – a discriminator network and a generator network by eliminating the fully connected layer in the traditional network and applying batch normalization and deconvolution operations. This paper also uses a hyper-parameter to measure the diversity and quality of the generated image. The experimental results of the model on the CelebA dataset show that the model has excellent performance in face image generation. In this report we take this approach to an entirely new type of style, I.e hand drawn animation. This is inherently challenging due to the fact that CelebA dataset contains data of similar style, real world palette. But, In terms of anime, the art style differs and palette is volatile at best.
With the increasing use of deep learning in the field of computer vision, automatic image generation based on depth models has received more and more attention. By using the powerful learning ability of the depth model, the inherent distribution law in the data can be efficiently mined to generate images with similar distribution.High-quality generated images can also be used to expand the amount of image data in the dataset, which can alleviate the need for a large number of training samples during deep learning model training, so that practical applications such as face recognition have higher accuracy.
The main objective of developing this project is
- To develop a Architecture that is able to handle a Volatile pattern of images and able to successfully generate new images based on learned latent distribution.
- To determine the faults and mode collapse conditions which may lead to breakdown of GAN