You are on page 1of 29

Academic Year 2023 - 24

S8 PROJECT WORK II
Third Review
PROJECT TITLE
AI BASED PERSONALIZED OUTFIT RECOMMENDER

BIP PROJECT ID CATERGORY


24S8EOW442 EXTERNAL-OWN

BATCH MEMBERS GUIDE


SURYA S S (202IT242) SELVAKUMAR M
SAMYUKTHA K R P (202IT210) Assistant Professor
Department of Information Technology
LITERATURE SURVEY

Works carried out


( with details of Methods/
Journal Paper Title with Materials/ Software/ Information gathered relevant to your
Sl.No.
Author Algorithms / fabrication / project
techniques/ components
used )
Personalized fashion Transfer learning from This paper a personalized Fashion
recommender system with ResNet50 Recommender system that generates
image based neural recommendations for the user based on
1
networks.- M Sridevi , N an input given.
ManikyaArun

Privacy Preserving Cloth Mobile Augmented Reality This paper proposes an alternative
Try-On Using Mobile approach that allows a customer to
Augmented Reality watch a three-dimensional (3D) model
2
Yoones A. Sekhavat of her/him wearing garments on a
personal mobile device using augmented
reality (AR)..

23-03-2024 THIRD REVIEW PRESENTATION 2


LITERATURE SURVEY

Machine Learning Models with Machine Learning In this paper, we tackle the latter issue, and perform
Optimization for Clothing experimental analysis of the various Machine
Recommendation from Personal Learning techniques that can be used for carrying out
3 Wardrobe the task. Since the recommendations must be made
Mahir Jain; Suraj Singh; K. from a user's personal wardrobe, the recommender
Chandrasekaran system doesn't follow a traditional approach.
Fashion Images Classification Machine Learning, Fashion is the way we present ourselves which mainly
using Machine Learning, Deep Deep Learning and focuses on vision, has attracted great interest from
4 Learning and Transfer Learning Transfer Learning computer vision researchers. It is generally used to
Models search fashion products in online shopping malls to
know the descriptive information of the product.
An Online Recommendation Deep Learning and In this paper, a deep learning-based online
System Using Deep Learning for Convolutional recommendation system has been proposed with a
Textile Products. Neural Network Convolutional Neural Network (CNN). Classes of
Ümit Turkut; Adem Tuncer (CNN). different patterns in the CNN architecture have been
5
determined according to users' and designers' pattern
preferences. The deep learning model recommends
patterns considering color compatibility for textile
products.

23-03-2024 THIRD REVIEW PRESENTATION 3


AIM & OBJECTIVES OF THE PROJECT
(Problem Statement)

Aim:

To develop an AI-driven Personalized Outfit Recommendation system.

Objectives:
• To implement Convolutional Neural Network (CNN) algorithms for in-depth analysis of
user preferences and fashion trends.
• To integrate real-time weather data to provide weather-appropriate clothing
recommendations.
• To improve the user experience by offering a sophisticated and personalized approach to
dressing for various occasions.
• To implement try on feature in which user can upload their image and desired cloth image
to see how that dress is suited on them.
23-03-2024 THIRD REVIEW PRESENTATION 4
SCOPE OF THE PROJECT

• Provide personalized outfit recommendations based on user preferences and real-time


factors.
• Event-specific Recommendation in tailor suggestions for different occasions, including
weddings, parties, and travel.
• To incorporate weather conditions to recommend suitable clothing options.

• The design an intuitive interface for effortless exploration and selection of outfits.

23-03-2024 THIRD REVIEW PRESENTATION 5


NEED FOR THE CURRENT STUDY

• The integration of AI in recommending personalized outfits aligns with the trend towards
tailored and individualized services.
• In particularly technological Convolutional Neural Networks, presents an opportunity to
create a sophisticated and efficient personalized outfit recommender.
• The rapid evolution of fashion trends necessitates an intelligent system that can keep pace
with the dynamic nature of the industry.
• In the recognizing the significant influence of weather conditions on outfit choices, there
is a need to integrate real-time weather data for more contextually relevant
recommendations.

23-03-2024 THIRD REVIEW PRESENTATION 6


FEASIBILITY ANALYSIS

Technical Feasibility:
• The required technology, such as Convolutional Neural Networks and image recognition,
is readily available and feasible for implementation.
• Skilled developers with expertise in AI and image processing are accessible.

Financial Feasibility:
• Budget considerations are within a reasonable range for developing and maintaining the
AI-based platform.
• Potential return on investment is justifiable given the demand for personalized fashion
solutions.

23-03-2024 THIRD REVIEW PRESENTATION 7


FEASIBILITY ANALYSIS Cont….

Operational Feasibility:

• The system's integration with real-time weather data and continuous learning algorithms
aligns with operational requirements.

• User-friendly interfaces and intuitive design contribute to operational ease.

Market Feasibility:

• Market research indicates a demand for personalized outfit recommendations, making


the project commercially viable.

• Identifying a niche in the market contributes to the feasibility of the proposed AI-driven
platform.

23-03-2024 THIRD REVIEW PRESENTATION 8


CHOICE OF COMPONENTS / MODULES / METHODS/TECHNIQUES
EQUIPMENT USED FOR PROJECT DEVELOPMENT
MODULE DESCRIPTION

Dataset Collection:

This involves gathering a diverse and representative dataset of fashion images. The dataset
should cover various styles, occasions, and weather conditions to train the CNN effectively.

Image Pre-processing:

Prior to training, images undergo pre-processing to standardize dimensions, enhance quality,


and normalize color schemes. This step ensures uniformity in the dataset, optimizing the CNN's
learning process.

Image Segmentation:

Image segmentation involves dividing images into meaningful segments or regions, such as
clothing items. This segmentation aids in isolating specific features during the CNN's training,
contributing to more accurate recommendations.
23-03-2024 THIRD REVIEW PRESENTATION 9
CHOICE OF COMPONENTS / MODULES / METHODS/TECHNIQUES
EQUIPMENT USED FOR PROJECT DEVELOPMENT
Training the CNN Algorithm:

In this pre-processed and segmented images are used to train the Convolutional Neural
Network. The CNN learns to recognize patterns, textures, and features associated with user
preferences, event types, and weather conditions.

User Interface:

The User Interface (UI) module focuses on designing an intuitive and user-friendly interface.
Users interact with the system through this interface to input preferences, view
recommendations, and engage with outfit try-on features.

23-03-2024 THIRD REVIEW PRESENTATION 10


CHOICE OF COMPONENTS / MODULES / METHODS/TECHNIQUES
EQUIPMENT USED FOR PROJECT DEVELOPMENT
Outfit Recommendation:

Based on the trained CNN model, this module generates personalized outfit
recommendations considering the user's preferences, the type of event, and real-time weather
conditions.

Try on the Outfit (Computer Vision):

In the computer vision technology, this module allows users to virtually try on the
recommended outfits. It asks to upload the cloth images and user image, then it will process
the images and will display the cloth image overlayed on the user.

23-03-2024 THIRD REVIEW PRESENTATION 11


PROJECT WORK PLAN

1. Project Initiation

2. Research and Planning

3. Data collection and Preprocessing

4. Model Development

5. Frontend Development

6. Backend Development

7. Weather API Integration

8. Try On model Development

9. Integration and Deployment

10. Documentation
23-03-2024 THIRD REVIEW PRESENTATION 12
EFFECTIVE UTILIZATION OF THE MODERN TOOL

Front End & Backend:


• HTML,CSS,JS, REACT for Frontend.
• Flask for Backend.
• The database used is firebase real-time database, LiveData, Firebase auth.

Outfit Recommendation Model:


• CNN & RESNET 50.
• Dataset: Myntra Fashion Dataset with nearly 15k records.

Weather API & Try On:


• OpenWeatherMap API & OpenCage API for real time weather forecast .
• Detectron2, OpenCV, Graphonomy, Posenet model for Try On feature.

28-02-2024 THIRD REVIEW PRESENTATION 13


PROPOSED METHODOLOGY ( Flow Chart)

Start

Collect the dataset


in Fashion Admin

Image pre-process Feature extraction Image segmentation

Train the CNN


Split image
algorithm

Recommend the
User
Weather API Integration outfit based on real
interface
time weather

Try on computer
23-03-2024 visionPRESENTATION
Third REVIEW 14
PROPOSED METHODOLOGY ( Gantt Chart)
TASKS Week

1-2 3-4 5-6 7-8 8-9 9-10 11-15


TITLE
1
SELECTION
ANALYSIS OF
2
PROJECT

3 LITERATURE SURVEY

4 IMPLEMENTATION
OF THE PROJECT

5 MODEL
DEVELOPMENT

6 SOFTWARE
IMPLEMENTION

23-03-2024 THIRD REVIEW PRESENTATION 15


MODULES

1. DATA COLLECTION AND PREPROCESSING


• Collection of fashion dataset – Myntra Fashion Dataset from Kaggle
• Data preprocessing including normalization and resizing
• Splitting the data into training, validation and testing
2. DEEP LEARNING MODEL DEVLEOPMENT
• Selection of appropriate deep learning algorithm CNN (RESNET50)
• Model architecture design
• Training the model on the preprocessed dataset
3. Front End Development
• Develop the Front End of a Fashion Ecommerce with HTML,CSS, REACT
• Integration of the Weather API
• Integration of OpenAI API
• Host the Front End Part

23-03-2024 THIRD REVIEW PRESENTATION 16


MODULES

4. Back End Development


• Develop backend using Flask
• Integrate it with the trained RESNET50 model.
5. Try On Development
• Develop the Try On mode with Detectron2, OpenCV, Graphonomy, Posenet
model .
• Performance evaluation and fine tuning.
6. TESTING AND VALIDATION
• Functional testing of outfit recommendation and other features
• Usability testing with potential users

23-03-2024 THIRD REVIEW PRESENTATION 17


COST BENEFIT ANALYSIS

• Given the utilization of open-source technologies, the project requires zero rupees in terms of
software costs.

23-03-2024 THIRD REVIEW PRESENTATION 18


RESULT AND DISCUSSION

ResNet 50 Model Performance:


ResNet50 and VGG16 pretrained models extract features from a dataset of Myntra
product images. Images are resized, preprocessed, then passed through the models to capture
visual information for tasks like classification and similarity comparison, facilitating efficient
analysis and understanding of large image collections.
Hybrid Model Performance Analysis:
This Model processes a Myntra dataset for a recommendation system. It cleans data, explores
features, and implements recommendation techniques like popularity-based and collaborative
filtering. Preprocessing involves handling null values and text preprocessing. Recommendations
are made based on user behavior, utilizing TF-IDF vectorization and cosine similarity for
content-based suggestions.

23-03-2024 THIRD REVIEW PRESENTATION 19


RESULT AND DISCUSSION

Weather Based Suggestion:


The weather component of the front end will give outfit suggestion based on real time weather .
This gets the user’s location by GPS (Location Access) and then with the help of
OpenWeatherMap API it gets the user’s location’s weather forecast.
Occasion Based Suggestion:
Users input an occasion type, and the chatbot suggests outfits accordingly. The system
processes user prompts, recommending suitable attire for various occasions, enhancing user
experience through personalized outfit suggestions based on input preferences.

23-03-2024 THIRD REVIEW PRESENTATION 20


RESULT AND DISCUSSION

Try On Performance:
In the front end user need to upload two images, one is cloth image and other is person image.
The model which is integrated in the backend will process the user uploaded images and will
give the final image i.e., cloth overlayed on user will be displayed in the frontend.
DISCUSSION:
Future Enhancements:
The findings from this chapter lay the groundwork for future enhancements. Areas such as
improved recommendation algorithm, integration of social media, Live virtual try on feature,
user feedback and rating system.

23-03-2024 THIRD REVIEW PRESENTATION 21


CONCLUSION

In conclusion, the project "AI-based personalized outfit recommender” based on user

preference, real-time weather, and try-on , offers numerous benefits such as

tailored outfit suggestions and real-time weather integration. However, challenges include

technical complexity, privacy concerns, potential inaccuracies, and dependency on external

APIs. Success hinges on effective implementation, addressing privacy issues, and refining

algorithms to ensure accurate recommendations and engaging user experiences.

23-03-2024 THIRD REVIEW PRESENTATION 22


INDIVIDUAL CONTRIBUTIONS TO THE WORK
Batch Member 1 : (202IT242 & SURYA S S)
1. User Preference Recommendation Model.
2. Try on Development.
3. Back End Development.
4. Integration of webpage with the Trained Models.
5. Model Performance Evaluation.

23-03-2024 THIRD REVIEW PRESENTATION 23


INDIVIDUAL CONTRIBUTIONS TO THE WORK
Batch Member 2 : ( 202IT210 & SAMYUKTHA K R P)
1. Dataset Collection & Preparation.
2. Webpage Development.
3. Weather API & OpenAI API Integration.
4. Integrate Front End with Back End.
5. Documentation and Report Preparation.

23-03-2024 THIRD REVIEW PRESENTATION 24


Status on Partial Completion and Submission of Project Report
List of Documents to be Submitted
SL.No List of Documents Status ( Provide the drive link
of prepared document)

1
Cover Page & Title Page (Both are in Cover & Title page
same format)
2 Bonafide Certificate Bonafide Certificate

3 Declaration Declaration

4 Acknowledgement Acknowledgement

5 Chapter I – Introduction Introduction

6 Chapter 2 – Literature Survey Literature Survey

23-03-2024 THIRD REVIEW PRESENTATION 25


Submission of Project Report
List of Documents to be Submitted
SL.No List of Documents Status ( Provide the drive link
of prepared document)
7 Chapters 3 - Objectives and Methodology Objectives & Methodology

Chapters 4 - Proposed work modules Proposed work modules


8
(Chapter name can be based on the work)
9 Chapter 5 - Results and Discussion Results & Discussion
10 Chapters 6 - Conclusions & Suggestions Conclusions & Suggestions
for Future work
11 References References

12 Appendices Appendices

23-03-2024 THIRD REVIEW PRESENTATION 26


PUBLICATIONS (CONFERENCE / JOURNAL / PATENT)

SL.No Paper/Manuscript Title Status ( Provide the drive


link of prepared document)
1
AI BASED PERSONALIZED OUTFIT RECOMMENDER Certificate

23-03-2024 THIRD REVIEW PRESENTATION 27


ORIGINALITY SCORE

SL.No Tool name for checking originality score Status ( Provide the drive
link of prepared document)
1
TURN IT IN Plagiarism

23-03-2023 THIRD REVIEW PRESENTATION 28


REFERENCES
( Journal Papers/ Books/ Website References )
1. C.-W. Hsieh, C.-Y. Chen, C.-L. Chou, H.-H. Shuai and W.-H. Cheng, "Fit-me: Image-based virtual try-on
with arbitrary poses", 2019 IEEE International Conference on Image Processing (ICIP), pp. 4694-4698, 2019.

2. N. Zheng, X. Song, Z. Chen, L. Hu, D. Cao and L. Nie, "Virtually trying on new clothing with arbitrary
poses", Proceedings of the 27th ACM International Conference on Multimedia, pp. 266-274, 2019.

3. Y. Lin, P. Ren, Z. Chen, Z. Ren, J. Ma and M. De Rijke, "Explainable outfit recommendation with joint outfit
matching and comment generation", IEEE Transactions on Knowledge and Data Engineering, vol. 32, no. 8,
pp. 1502-1516, 2019.

4. K.-H. Liu, H.-H. Liu, P.K. Chan, T.-J. Liu and S.-C. Pei, "Age estimation via fusion of depthwise separable
convolutional neural networks", 2018 IEEE International Workshop on Information Forensics and Security
(WIFS), pp. 1-8, 2018.

5. W. Zhou, P. Y. Mok, Y. Zhou, Y. Zhou, J. Shen, Q. Qu, et al., "Fashion recommendations through cross-media
information retrieval", J. Visual Commun. Image Represent, vol. 61, pp. 112-120, 2019.
23-03-2024 THIRD REVIEW PRESENTATION 29

You might also like