International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 11 Issue: 03 | Mar 2024
p-ISSN: 2395-0072
www.irjet.net
IMAGE GENERATION USING CP-VTON BY GENERATIVE ADVERSARIAL APPROACH B.Chandra Shekar1, M.Krishna Prasad2, K.Bhaskar Chowdary3, C.Sai Pavan4, S.Vinay Raj5 1Student, Dept. Of Computer Science Engineering, GITAM Deemed University, AP, INDIA 2Student, Dept. Of Computer Science Engineering, GITAM Deemed University, AP, INDIA
3Student, Dept. Of Computer Science Engineering, GITAM Deemed University, AP, INDIA
4Student, Dept. Of Computer Science Engineering, GITAM Deemed University, AP, INDIA
5Student, Dept. Of Computer Science Engineering, GITAM Deemed University, AP, INDIA
--------------------------------------------------------------------***-----------------------------------------------------------------move around and see the product from different Abstract
perspectives. This technology is especially useful for ecommerce companies, as it allows customers to get a better sense of what a product will look like on them before making a purchase, which can increase confidence in their buying decisions. Virtual try-on technology is becoming increasingly popular across many industries, from fashion and beauty to home decor and even automotive, as it provides a more immersive and personalized shopping experience for customers.
Recently, a visual experiment has become popular, the purpose of which is to project the image of the desired outfit onto a reference figure. Past artworks usually focus on maintaining the personality of the clothing image when it is transformed into any human pose (eg texture, logo, embroidery). However, creating photorealistic test images proves difficult when the reference person has visible occlusions and human poses. The goal of this project is to develop a virtual shopping system that uses deep learning techniques to provide an immersive and realistic shopping experience. The system uses Gaussian Mixture Model (GMM) and Texture Orientation Matching (TOM) algorithms to model the garment and its texture, respectively. The GMM algorithm is used to segment the garment from the background and extract its shape and features. The TOM algorithm then matches the texture of the garment to the texture of the user's body using the learned feature representation. The resulting virtual fitting system provides an accurate picture of how the clothing product would look in the user's eyes, allowing them to make informed purchasing decisions. The system was tested on several different garments and showed promising results in terms of accuracy and realism.
2. LITERATURE REVIEW In our project, we aim to leverage recent advancements in deep learning to enhance the virtual try-on experience. Just as advancements in deep learning have revolutionized medical research, our project seeks to push the boundaries of virtual try-on technology, improving the online shopping experience for users. In their study, Han et al. (2018): Viton: An Image-based Virtual Try-on Network. Our project could improve upon Viton by potentially offering more efficient training algorithms, faster inference times, or enhanced realism in virtual try-on results. Similarly, Song et al. (2019): Attentive Generative Adversarial Network for Virtual Try-on. Our project might surpass this model by introducing additional attention mechanisms or novel architectures that further improve clothing realism and user experience.
Keywords: Virtual Try-on, GMM, TOM, GAN
1.INTRODUCTION Deep learning is a branch of machine learning that uses artificial neural networks to process and analyze data. Neural networks consist of multiple layers of interconnected nodes or neurons that are able to learn and extract features from input data. Virtual Try-on is a technology that allows users to virtually try on products, such as clothing, accessories, eyeglasses, and makeup, using augmented reality (AR) or virtual reality (VR) technology. With virtual try-on, users can see how a product would look on them without physically trying it on. Virtual try-on works by using a camera to capture an image of the user, which is then overlaid with a 3D model of the product they want to try on. The user can then see how the product looks on them from different angles,
© 2024, IRJET
|
Impact Factor value: 8.226
Liu et al. (2020): Virtual Try-on with Detail-enhanced Feature Transformation. Our project could advance beyond this approach by integrating more sophisticated feature transformation techniques or incorporating additional details such as fabric texture or wrinkles, leading to even more realistic virtual try-on results. Similarly, our project aligns with the goals of the studies conducted by Xie et al. [4], and Li et al. [5], which focus on enhancing virtual try-on realism through various techniques such as attention mechanisms, detail-enhanced feature transformation, large-scale variational autoencoders, and GAN-based approaches, respectively.
|
ISO 9001:2008 Certified Journal
|
Page 1487