1 of 27

STYLE

BUSTERS

Team 12

Clothes

Logos

Artworks

2 of 27

INTRODUCTION

We introduce a new searching technique for finding similar style Clothes, Artworks, and Logos.

3 of 27

PROBLEM DEFINITION

FASHION DESIGNERS CANNOT FIND OUT IF THEIR DESIGNS HAVE BEEN STOLEN

2

1

SHOPPERS CANNOT FIND SIMILAR STYLE CLOTHES FROM VARIOUS E-COMMERCE WEBSITES

4 of 27

PROBLEM DEFINITION

THERE ARE CURRENTLY NO RELIABLE METHOD TO FIND SIMILAR LOGOS

Looka.com

FreeLogoDesign.org

LogoMaker.com

5 of 27

PROBLEM DEFINITION

THERE ARE CURRENTLY NO RELIABLE METHOD TO FIND SIMILAR STYLE ARTWORKS

6 of 27

OBJECTIVES

CLOTHES

Our system shall allow the users to search for similar style clothes.

RELIABLE SEARCH

Our system introduces a new searching technique for reliable search.

LOGOS

Our system shall allow the users to search for identical logos.

ARTWORKS

Our system shall allow the users to search for similar style artworks.

7 of 27

DATASETS DESCRIPTION

8 of 27

WIKIART DATASET

Size: 25.5 GB

Number of classes: 23 different artists and 27 different styles Number of images: 80000

9 of 27

DEEPFASHION2 DATASET

Size: 13 GB

Number of classes: 13 different clothing categories Number of images: 491K diverse image

10 of 27

FASHION PRODUCT IMAGES DATASET

Size: 25 GB

Number of classes: 28 Number of images: 44K

11 of 27

CLOTHES SYSTEM OVERVIEW

12 of 27

LOGOS SYSTEM OVERVIEW

13 of 27

ARTWORKS SYSTEM OVERVIEW

14 of 27

Introducing a new search technique

They Search Locally Very High Storage Cost

Must be updated Regulary

They Search In Random Sources

They Cannot Control Accuracy They Display Limited Results

Not Reliable For Professinal Domains Faster

Traditional Method

Our Method

We Search Remotely Low Storage Cost

No Updates Required

We Only Search In Reliable Sources We Can Control Accuracy

We Display All The Results

Can Be Used In Professional Domains A bit Slower

15 of 27

CLASS DIAGRAM

16 of 27

Clean Architecture

DESIGN PATTERN

17 of 27

SEQUENCE DIAGRAM

Server

Feature extraction

Application

18 of 27

USER INTERFACE

HomePage

No Connection

Upload clothes Image

Retrive image of Clothes

19 of 27

USER INTERFACE

Upload artworks image

Artworks results email

Upload Logo Image

Logos results email

20 of 27

Artworks

RESULTS

VGG 11

Training accuracy:- Validation accuracy :-

90%

77.0377%

The results shown above were achieved using the Wiki-Art dataset and VGG 11 model.

Training accuracy:- 89.71%

validation accuracy:- 76.6696%

The results shown above were achieved using the Wiki-Art dataset and VGG 16 model.

21 of 27

CLOTHES OBJECT DETECTION

CONFUSION MATRIX

22 of 27

LOGOS DEMO

23 of 27

ARTWORKS WEBSCRAPING DEMO

24 of 27

ARTWORKS FEATURES EXTRACTION DEMO

25 of 27

CLOTHES OBJECT DETECTION

DEMO

26 of 27

WHAT'S NEXT

1

2

Training a Feature extractor to extract the style of clothes

Developing a mobile application to link all of our system components together

3

Developing a website to go alongside the mobile application

27 of 27

THANK YOU