Fashion Recommendation System
Using Deep Learning
Guided By – Dr. Shaik Subhani
Team:
K.Subash – 22MIS7139
Shaik Abdul Kalam – 22MIS7089
D.Naveen Akash – 22MIS7143
Zeroth Review
Title of the Project Fashion Product Image
Recommendation System
Operating System Server: Windows 10
or later / macOS / Linux
Database Server: None (feature storage
in local filesystem; can be extended with
SQLite/MySQL)
Software Requirements
Client: Any modern web browser
(Chrome, Edge, Firefox)
Tools: Visual Studio Code, Jupyter
Notebook (optional)
User Interface: Streamlit Web App
Code Behind: Python 3.x (TensorFlow,
Keras, Pillow, Scikit-learn, Streamlit)
Processor: Intel Core i3 or higher /
Equivalent AMD
Hardware Requirements RAM: 4 GB minimum, 8 GB or higher
recommended
Hard Disk: PC with at least 20 GB free
space (depending on image dataset size)
GPU: Optional but recommended for
faster deep learning operations
Abstract
This project is aimed at developing a web-based Fashion
Product Recommendation Tool focused on providing visually
similar product suggestions to users based on uploaded
clothing images. The system uses deep learning to
automatically extract features from fashion images, compares
them, and then recommends products from a catalog that
closely match the style and appearance of the uploaded item.
This tool can be used by online retailers to enhance product
discovery, by stylists to curate collections, or by individual users
looking for inspiration or alternatives to a favorite fashion piece.
Functional components of the project:
There are two main user types in the system: casual users
(shoppers) and catalog managers (admin). Users interact via a
simple web interface, upload an image of any clothing item, and
instantly receive a list of the most visually similar items from
the inventory. Catalog managers can update or expand the
reference dataset.
The modules involved are:
Product Catalog Management
Image Preprocessing
Feature Extraction
Recommendation Engine
Search & Reports
Authentication (optional/simple)
Product Catalog Management:
In this module, the administrator or catalog manager uploads a
collection of product images to build the reference database.
The system preprocesses and organizes these images for
efficient access and searching.
Image Preprocessing:
Here, all input images (product and user-uploaded) are resized,
normalized, and prepared for feature extraction. This ensures
consistent quality and computational efficiency for deep learning
tasks.
Feature Extraction:
A pre-trained Convolutional Neural Network (CNN) model (e.g.,
VGG16) is used to extract meaningful numerical features from
each product image. These features are stored for similarity
comparison.
Recommendation Engine:
When a user uploads an image, the same feature extraction
process runs, and the system then computes similarity scores
against all catalog items. The top matches are recommended to
the user visually within the app.
Search & Reports:
This module enables searching the product catalog by attributes
(category, tags if available) or by visual similarity. Reports can
include popular queries, frequent matches, or catalog coverage.
Authentication:
Basic authentication can restrict the management functions
to authorized users only. Standard users (shoppers) may
browse and upload without login.