lit

Download as pdf or txt
Download as pdf or txt
You are on page 1of 5

SREE NARAYANA MANGALAM

INSTITUTE OF MANAGEMENT & TECHNOLOGY


MALIANKARA P.O., N. PARAVUR, ERNAKULAM - 683516

DEPARTMENT OF
COMPUTER SCIENCE AND ENGINEERING

LITERATURE SURVEY

Group 2
CROSS-SITE VIRTUAL APPAREL TRY-ON

CSE 2024-2025

Project Team Members


1. Afna Iqbal, Roll No.-4
2. Nandana Linson, Roll No.-34
3. Rithika C B, Roll No.-36

Guided by: Mrs.Archana K S

Date of Submission :
LITERATURE SURVEY

Enhancing the Virtual Jewelry Try-On Experience With


Computer Vision

The literature survey in the paper "Enhancing the Virtual Jewelry Try-On Experience
with Computer Vision" reviews the progression of virtual try-on systems, highlighting
the role of computer vision in transforming online jewelry shopping. Early virtual try-
on methods primarily relied on 2D image overlays and basic 3D modeling, which,
while functional, lacked realism and user engagement, limiting widespread adoption.
Later, the integration of 3D body scanning and modeling improved accuracy and
visual appeal, allowing users to experience a better fit, though these methods were
resource-intensive and complex to implement. With advancements in Augmented
Reality (AR) and Virtual Reality (VR), retailers such as Magic Mirror and Warby
Parker introduced more interactive platforms, enabling customers to try on products
in virtual environments. However, these systems still faced limitations such as high
costs and technical complexities. In recent years, computer vision-based techniques
have emerged as a more accessible solution, using image recognition and body pose
estimation to enhance realism and interactivity without the need for intricate 3D
modeling. Major retailers like Amazon, Zara, and Tiffany & Co. have adopted such
virtual try-on technologies, though challenges like lighting conditions and privacy
issues persist. The paper aims to address these shortcomings by presenting a cost-
effective, computer vision-driven virtual try-on system using Google’s Mediapipe
library, focusing on rings, earrings, and bracelets, and offering a scalable and realistic
solution for both customers and jewelers.The proposed virtual jewelry try-on system
is cost-effective, using computer vision and Python libraries like Mediapipe and
OpenCV, avoiding expensive 3D modeling. It offers realism, accuracy, and user-
friendliness, allowing easy at-home try-ons. Scalable for future categories, it
eliminates size concerns, enhancing online jewelry shopping for customers and
jewelers.The proposed system has limitations, including sensitivity to lighting and
occlusion, privacy concerns with real-time image capture, and reliance on camera
quality for accurate results. It currently supports only rings, earrings, and bracelets,
lacking other jewelry categories and the depth of realism provided by 3D modeling in
AR/VR systems.
VITON: An Image-based Virtual Try-on Network

The literature review in the paper “VITON: An Image-Based Virtual Try-On


Network” explores the evolution of virtual try-on systems, with an emphasis on
fashion analysis and image synthesis techniques. Early work in fashion analysis
mainly focused on clothing recognition, attribute matching, and fashion trend
prediction. These methods were useful for identifying and recommending fashion
items but did not tackle the complex task of virtually trying on clothes. Generative
Adversarial Networks (GANs) have made significant strides in image generation and
editing tasks, including image-to-image translation, where an input image is
transformed into another form, such as creating an RGB image from an edge map.
However, existing GAN-based approaches are limited in handling geometric changes
and preserving the intricate details necessary for virtual try-on, such as how clothing
deforms on various body shapes and poses. Traditional virtual try-on systems often
relied on 3D measurements and complex hardware, making them expensive and
difficult to deploy on a large scale. The VITON framework aims to overcome these
challenges by synthesizing a photo-realistic image of a person wearing a target
clothing item using only 2D images. It introduces a coarse-to-fine framework that
preserves the body’s pose and shape while ensuring detailed visual patterns of the
clothing are retained. The VITON paper offers a cost-effective and scalable virtual
try-on solution using only 2D images, eliminating the need for expensive 3D modeling
and hardware. Its coarse-to-fine framework preserves body pose and shape while
ensuring realistic deformations and detailed visual patterns of clothing, enhancing the
user experience.The VITON system has limitations, including difficulty handling rare
or complex poses, occasional artifacts around clothing regions like necklines, and
challenges in seamlessly integrating garments with significantly different shapes.
Additionally, its reliance on 2D images may lack the depth and realism achievable
with 3D modeling techniques.

Deep Learning in Virtual Try-On: A Comprehensive Survey


The paper "Deep Learning in Virtual Try-On: A Comprehensive Survey" provides an
extensive review of deep learning models used in virtual try-on systems. It highlights
the increasing relevance of virtual try-on technology in transforming online shopping
by allowing customers to visualize clothing and accessories digitally. This technology
addresses major online retail challenges, such as uncertainty about fit and style,
improving customer satisfaction and reducing return rates. The review categorizes
virtual try-on models into three types: image-based, multi-pose, and video-based,
offering detailed technical summaries and performance comparisons. The paper
emphasizes the use of Generative Adversarial Networks (GANs) and diffusion models
for image synthesis and highlights their strengths and limitations, such as difficulties
in maintaining clothing textures and handling significant spatial deformations. It also
discusses the datasets used to train these models, noting biases like overrepresentation
of female models and simple backgrounds. Additionally, the survey explores the
impact of virtual try-ons on customer satisfaction and business outcomes, including
increased conversion rates and reduced return rates. However, challenges like
preserving facial identity, dataset diversity, and economic impact are noted, pointing
to future research directions for improving accuracy, realism, and applicability of
these models. This comprehensive survey fills a literature gap by analyzing deep
learning-based virtual try-on technologies and their broader implications for e-
commerce.The paper “Deep Learning in Virtual Try-On: A Comprehensive Survey”
provides a thorough analysis of deep learning-based virtual try-on models,
highlighting their potential to revolutionize online shopping by enhancing customer
satisfaction and reducing return rates. It categorizes models into image-based, multi-
pose, and video-based, offering in-depth technical comparisons and performance
metrics. The paper addresses challenges like dataset biases and limitations in clothing
texture preservation while identifying future research directions, making it a valuable
resource for advancing virtual try-on technology in e-commerce. The paper “Deep
Learning in Virtual Try-On: A Comprehensive Survey” has some limitations,
including a lack of real-world economic impact analysis, as it doesn’t provide concrete
data on how virtual try-ons affect business performance. It also highlights dataset
biases, such as overrepresentation of female models and limited diversity in clothing
types and body shapes, which can lead to model inaccuracies. Additionally, challenges
in preserving facial identity and handling complex textures or spatial deformations in
try-on images remain unresolved.

A Virtual Try-on System Based on Deep Learning


The paper "A Virtual Try-on System Based on Deep Learning" presents a system
designed to enable users to virtually experiment with clothing using advanced deep
learning technologies. The authors utilize generative adversarial networks (GANs)
and style transfer algorithms to create visually accurate representations of how
garments would appear on users. The system first predicts the semantic layout of the
reference image, which helps maintain the fine details of the clothing. The try-on is
then completed in a virtual environment, generating the final output. The system
architecture consists of front-end and back-end components, with the latter performing
model computations through layers of data processing and algorithmic design. The
system is optimized using ACGPN (Adaptive Content Generating Network) to handle
complex scenarios such as occlusions involving limbs. Extensive datasets, including
CP-VTON and VITON, are used for training the model, ensuring high accuracy. The
experimental results show that the system offers efficient, high-quality virtual try-on
experiences. The authors acknowledge certain limitations, such as challenges in
dealing with complex clothing textures and patterns, but suggest further improvements
in scalability and efficiency for future research. This work holds promise for
enhancing e-commerce platforms by improving user experiences and reducing the cost
associated with traditional fitting methods.A virtual try-on system based on deep
learning offers several advantages. It enhances the shopping experience by allowing
users to visualize how clothes will look without physically trying them on. This saves
time and effort for both consumers and businesses. The system improves accuracy and
realism using advanced algorithms like Generative Adversarial Networks (GANs),
ensuring high-quality visual results. Additionally, it reduces operational costs by
eliminating the need for physical inventories, making it highly scalable for e-
commerce applications Despite its advantages, a virtual try-on system based on deep
learning has some disadvantages. It requires high-quality input images for accurate
results, and complex clothing textures or patterns can be challenging to handle.
Additionally, the system’s reliance on deep learning models means significant
computational power is necessary, which may lead to higher processing times. There
are also limitations in handling real-world variations, such as different body poses or
clothing occlusions, which can reduce accuracy.

You might also like