International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 10 Issue: 05 | May 2023
p-ISSN: 2395-0072
www.irjet.net
TEXT TO IMAGE GENERATION USING GAN Mrs. L. Indira1, M. Sunil2, M. Vamshidhar3, S. Ravi Teja4, R. V. Praneeth5 1 Assistant Professor, Department of Computer Science and Engineering
2,3,4,5 Undergraduate Student, Department of Computer Science and Engineering, Vallurupalli Nageswara Rao
Vignana Jyothi Institute of Engineering and Technology, Hyderabad, Telangana, India ---------------------------------------------------------------------***--------------------------------------------------------------------Abstract - Producing good images from descriptions is a challenge in computer vision with practical applications. To address this
issue, we propose Stacked Generative Interconnected Networks (StackGAN) to combine the 256×256 real images described in the annotation. The process splits the problem into two phases: Phase-I generates the low-level image by drawing pictures and colours from the text, while Phase-II edits the Phase-I results to create the high image with photorealistic details. . price picture. Conditional magnification is used to improve image contrast and stable GAN training. Extensive testing demonstrates that our development method is better than the state-of-the-art method and proves its effectiveness in creating realistic images based on descriptions. In summary, StackGAN provides a multi-level approach with additional optimization to improve composite images and shows great results in creating high-quality images from text. Key Words: Stack GAN, Resolution, Conditioning Augmentation, Image generation
1.INTRODUCTION Creating realistic images from descriptions is an important and difficult task, applicable in many fields such as photo editing and computer aided design. Recently, prolific competing networks (GANs) are showing promise in creating real images. Conditional GANs specially designed to generate images from descriptive text have been shown to be able to generate textrelated images. However, training GANs to create realistic images from annotations remains challenging. Increasing the number of layers used to solve image problems in current GAN models often makes training unstable and ineffective. The problem arises because the classification of the natural image and the implicit classification model do not overlap well in high pixel space. This problem becomes more serious as the image resolution increases. Previous tutorials were limited to creating sensible 64×64 images from descriptions without real objects and details like a bird's beak and eyes. Also, annotations are required to create higher resolution images such as 128×128. To solve these problems, we present Stacked Generative Adversarial Networks (StackGAN), which decomposes the problem of text-to photorealism image synthesis into two manageable problems. We first developed the decoder using a level-I GAN. Next, we set the Level-II GAN on top of the Level-I GAN to create a really good image (for example, 256 × 256) according to the LevelI results and description. Leveraging the Tier-I results and text, the Tier II GAN learns to capture information the Tier-I GAN would miss and add more detail to the product. This approach improves the rendering of high-resolution images by increasing the ability to model the distribution of the image distribution. We also propose developing a new method to handle the differences in various text events caused by the limitation of the number of training text-image pairs. This technique facilitates smoothing of the central cooling manifold by allowing small random perturbations. It improves the contrast of synthetic images and improves the training of GANs. Our contributions can be summarized as follows: (1) We propose a common crosslinking algorithm that improves the actual (256 × 256 resolution) efficiency of image binding without decomposing the problem into control problems. (2) We introduce the amplification technique leading to different design and stability for training GANs. (3) Through extensive and thorough testing, we demonstrate the effectiveness of the overall design and the impact of individual products. Such useful information could guide the GAN design pattern in the future.
© 2023, IRJET
|
Impact Factor value: 8.226
|
ISO 9001:2008 Certified Journal
|
Page 1479