You are on page 1of 5

Convoluted Neural Network Test for CIFAR-10

Name – TUHIN KARAK


Entry no – 2023EEN2753
Date – 19th January, 2024
1. Load CIFAR-10 code with TensorFlow, matplotlib and NumPy import

2. Normalizing the Image Pixels from 0 to 1.


As the CIFAR-10 consists of 32x32x3 pixel size, i.e. its each pixel is of RGB, so the
image matrix is 3 dimensional.
So, each pixel will be having
R(Red) ranging from 0-255
G(Green) ranging from 0-255
B(Blue) ranging from 0-255
So, divided by 255.
3. Convolution layers and Dense layers

Consists of: -

i) Convolution Layer of 5 filters, filter size=5x5, strides=2, and activation


function=Relu.
ii) Max Pooling Layer of size 2x2, strides=2.
iii) Convolution Layer of 10 filters, filter size=3x3, strides=2, and activation
function=Relu.
iv) Max Pooling Layer of size 2x2, strides=1.
v) Dense layer of 256 neurons, activation=Relu
vi) Dense layer of 64 neurons, activation=Relu
vii) Output Dense layer of 10 neurons, activation=Relu as there are 10
classifications.
viii) Loss is Sparse categorical Crossentropy
ix) Optimizer is ADAM
x) Epochs = 20.

4. Train Accuracy is 54.66% and Test Accuracy is 51.17%, Train Loss is 1.2516
and Test Loss is 1.3714
5. Train Accuracy and Loss Graphs vs Epochs:

Train Accuracy Graph vs Epochs

Train Loss Graph vs Epochs


6. Test Accuracy and Loss Graphs vs Epochs

Test Accuracy Graph vs Epochs

Test Loss Graph vs Epochs


7. Model Summary

Model Summary

You might also like