Scribd
Upload
8 views6 pages

Sign Detection

The document outlines a project on handwritten digit recognition using the MNIST dataset and traffic sign detection using CNNs, detailing the steps involved from data import to model evaluation. It includes information on libraries used, data preprocessing, augmentation techniques, and model compilation. The final model aims to classify traffic sign images into 58 classes based on the trained dataset.

Uploaded by

preethikaburla
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
8 views6 pages

Sign Detection

The document outlines a project on handwritten digit recognition using the MNIST dataset and traffic sign detection using CNNs, detailing the steps involved from data import to model evaluation. It includes information on libraries used, data preprocessing, augmentation techniques, and model compilation. The final model aims to classify traffic sign images into 58 classes based on the trained dataset.

Uploaded by

preethikaburla
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 6

26-05-25

------------
MNIST Hand written digit
10 classes (0-9)
70000 hand written digits
60000- Train
10000- Test
CNN (Convolutional Neural Networks):
Resnet
Efficient Net
Google Net
convnet

Traffic sign detection using CNN: (58 classes)


40-50% Because of signaling

step1:
Import the libraries
pandas- Data frames /analysis
numpy- numerical python (arrays)
matplotlib- data visualization
sklearn-pre-defined ml and dl
libraries (modules)
tensorflow- frame work used to work
images in deep learning and machine learning
opencv- image processing and
handling

step2:
Load the dataset from kaggle
58 classes (multi class
classification problem)

step3:
Data exploration and Data
visualization:
All the images
All the labels of images
the dataset information
statistical attributes'
sample images:
There are 58 classes and
58 images
class id Name (labels)
53
Give way
55 No
entry
54 No
stop

step4:
Pre-processing (Data preparation):
Because the data may not be in the
required format
Hence we will pre-process into the
required format
Normalization
Scaling etc..

step5:
total dataset- 4170 images
data splitting 80:20
80% training- 3336
20% validation- 834

Train images, Train labels - used to


train the model
Validation images, validation labels-
evaluate the performance of the model
Test (unseen images)-the accuracy

step6:
When the data is limited and the
model may not perform well with this data.
Fine Tuning :
for example we have small number of
images
because of this we may get less
accuracy

Data Augmentation:
Its used to increase the dataset size
instead of adding the new data
it will transform the existing data by
applying various augmentation techniques
rotation
scaling
flipping
etc..
Model complilation:
model.compile(loss=tf.keras.losses.SparseCatego
ricalCrossentropy(),
optimizer='adam',
metrics=['accuracy'])

Testing:
you can give any traffic sign image as input
and our model will classify that image as one of
the class labels (out of 58 classes)

step7:
Build the model
CNN (Convolution, Pooling, Flatten,
SoftMax layer)
It will extracts the
features (important information) from the given
image

create the model:


Train the model:

step8:
Model evaluation:

You might also like