1
Practical Training Seminar
on
'DEEP LEARNING'
''Diabetic Retinopathy Detection''
Presented By:
Harsh Sharma
16EJCIT028
IT/7th Sem/A2
Submitted To:
1. Mr. Vishal Kumar Sagtani
( Seminar Coordinator , IT )
2. Mr. Piyush Gautam
( TPO, IT )
2
Contents
❑ Company Profile
❑ Introduction
❑ Introduction to Deep Learning
❑ Overview of Tools Used
❑ Diabetic Retinopathy Detection-
Description
3
Company Profile
Forsk Technologies is Jaipur based IT consulting
company which combine analytics expertise and AI
solutions to develop software, people and
organizations.
4
Company Profile
• The team consist of scientists and engineers —
that code and transform.
• Forsk Technologies provides its clients with a
variety of services. Some of them are :-
• Consulting agreement resources and
developing highly skilled resources on the field of
AI and Machine Learning for digital and business
transformation.
5
Introduction
• Today, along with the development of technologies, Data
Science applications have gained incredible success since
they allow us to share, analyze and sometimes even
predict information all over the world.
• Moreover, by providing highly functional and interactive
features, more and more businesses are now using Data
Science for a rapid growth.
• These applications are used for shopping, social networking,
banking and so on. One can even say it has unlimited uses.
6
Introduction
• According to a research Data has surpassed the value of OIL
in 2018 and is the most crucial resource in the coming
decade.
• This is because of the reason that more and more
information(DATA) is generated and collected everyday.
• However, this also has a lot of drawbacks as this data is not
always secured or stored with adequate precautions.
" There are certain mis-applications that can be done by gaining
access to this data and using for unethical predictions and
personal information.
7
DEEP LEARNING
• Data Science is a multi-discipilinary field that uses scientific
methods, processes, algorithms and systems to extract
knowledge and insights from Data.
• Data Science is a blend of various tools, algorithms, and
machine learning principles with the goal to discover hidden
pattern from the raw data.
8
DEEP LEARNING
• AI: Artificial intelligence (AI) is the simulation of human
intelligence processes by machines, especially computer systems.
• Machine Learning: It is a subset of Artificial Intelligence which
is used to perform a specific task without using explicit
instructions, relying on patterns and inferences.
9
DEEP LEARNING
• DEEP LEARNING: It is further the subset of machine
learning in artificial intelligence(AI) that has networks
capable of learning unsupervised from data that is
unstructured or unlabeled.
• Deep learning learns from vast amounts of unstructured
data that could normally take humans decades to
understand and process.
• It utilizes a hierarchical level of artificial neural networks
to carry out the process of machine learning.
10
Tools Used
1. Spyder: Anaconda
• Spyder is an IDE of Anaconda which is used for various python
libraries specially while working on machine learning.
Tools Used 11
2. Google Colab
• Google Colaboratory is a research tool for machine
learning education and research. It’s a Jupyter notebook
environment that requires no setup to use.
12
Libraries Used
1. NUMPY and PANDAS:
• Provides high-performance, easy to use structures and data
analysis tools.
• Provides fast mathematical computation on arrays and
matrices.
• Provides in-memory 2-d table object called 'Data frame'. It
is like a spreadsheet with column names and row labels.
• Provide SQL-like functionalities to filter, sort rows based
on conditions.
Libraries Used 13
2. KERAS:
• Keras is an open-source neural-network library written in Python. It is
capable of running on top of TensorFlow.
• From keras.applications I used VGG16, a pre-trained model
on convolutional neural network(CNN).
• From keras.preprocessing I used DataGenerator, which is used
for generating data in real time.
• From keras I also used optimizers, for calculating the loss in the model.
14
Diabetic Retinopathy Detection
• The main objective behind developing this customized
Diabetic Retinopathy Detection model is to reduce the human
error in classifying and predicting the stage of the blindness
caused by the diabetes.
• In this model , data has been collected from Kaggle , dataset
consists of 3662 retina images taken using fundus
photography under a variety of imaging conditions. These
images are classified into 5 classes( i.e, 5 different stages of
blindness) .
15
Diabetic Retinopathy Detection
• I have worked mainly on the Google colab for the training of
this model.
• I downloaded the data from kaggle directly to my drive using
kaggle library that saved me around 20 GB of data and hours
of downloading.
• I have used VGG16 ( a pretrained model ), on convolutional
neural networks(CNN) which is trained on a dataset of about
2 million images.
• This pretrained model consists of a number of
convolutional 2-d and maxpooling layers in a given order with
a particular number of nodes per layer.
16
Diabetic Retinopathy Detection
• I have also used ' ImageDataGenerator ' for realtime
generation of new images from the original by shearing,
rotating, zooming, fliping, rescaling etc. For increasing the
size of dataset for a better performance of the model.
• Then these images were used for the training of my model.
These images were generated and deleted at runtime saving
space and time by a huge margin.
• The model now operates at an accuracy of 72.61% on new
images given to model.
Diabetic Retinopathy Detection 17
Screenshots :
Downloading data from kaggle:
Diabetic Retinopathy Detection 18
Screenshots :
Preprocessing of data:
19
Diabetic Retinopathy Detection
Screenshots :
Using VGG16 :
20
Diabetic Retinopathy Detection
Screenshots :
Model Summary:
21
Diabetic Retinopathy Detection
Screenshots :
Inside VGG16:
22
Diabetic Retinopathy Detection
Screenshots :
Training and Accuracy:
23
Diabetic Retinopathy Detection
THANK YOU!!
