TEAM NAME: RESTYLE
TEAM DETAILS: NIDHI IYER
SAYALI KADAM
NITHYA SREE RAMIREDDY
Our Solution
Theme: Engagement on a shopping platform
Problem Statement: developing a user-friendly and accessible
recommendation system that analyzes a person's facial unique
features and suggests a personalized color palette for clothing
and other products.
Color analysis is a professional method for identifying colors that
harmonize with the user's natural coloring (skin tone, hair color,
eye color). Our goal is to recommend clothes that will actually look
good on the user and not just on the model by choosing color
palette that flatter each user. The 12 Seasons approach is a refined
system that builds upon the four-season approach by incorporating
additional factors like brightness and depth.
High Level Design
er
Us
er
lay
User takes Input survey Output
an input from user generated
image on myntra!
n
atio
plic
Ap er
lay
Face detection Color palette Product
Model generation recommendations
ase
tab
Da
er
lay
Face Fashion products
dataset dataset
Face Detection Model
Image Segmentation: User uploads an image. Using OpenCV and
the MediaPipe library, the selfie undergoes image segmentation and
creates a mask to isolate the face from the background. The
background is replaced with a default color.
Face Detection and Reshaping: OpenCV is utilized to detect the face in the
segmented image. The image is then reshaped to exclude non-facial parts,
ensuring accurate analysis.
Face Dataset: The dataset used to test this model is the Face dataset. A
collection of 7.2k+ images. A mix of all common creeds, races, age groups and
profiles to ensure no bias.
Color Palette Generation Model
RGB to HSL Conversion:The face image’s pixel colors are
converted from RGB to HSL to understand the hue(temperature),
depth and chroma of the image.
K-Means Clustering: The K-means clustering algorithm is applied to the
HSL values to identify distinct color clusters on the face. Each cluster
represents a dominant color.
Seasonal Mapping: Based on the detected temperature, depth, and chroma
levels, the model maps the user’s face to one of the predefined seasons. Each
season is associated with a specific color palette.
Product Recommendation Model
Taking user survey: The user selects the details of the
product which they are looking for. These selections act as
a filter to recommend the products the user wants.
Product Data Handling: The csv parser reads the available product data on the
website and filters out the product based on the user survey. The seasonal color
palette generated earlier further filters out clothes which are best for the user.
Fashion Product Dataset: It has 44k products with multiple
category labels, descriptions and high-resolution images.
Integration with myntra app: The user uploads a selfie, then we use face
detection and color palette generation to select a seasonal color palette which
works as a filter. This helps the user find products that complement their skin
tone, enhancing their shopping experience.