9 Capstone Project CNN - Participants Guide.ipynb

Kernel: Python 3 (ipykernel)

Objective

Build, train, and evaluate a CNN using Keras on CIFAR-10 To learn a model that correctly assigns each input image to exactly one of 10 predefined object classes.

CIFAR-10 (Canadian Institute For Advanced Research) is a benchmark image classification dataset widely used to evaluate Convolutional Neural Networks (CNNs) and other computer vision models.

It is designed to be:

• Small enough to train quickly

• Complex enough to demonstrate real-world image challenges

Data Composition

Attribute	Description
Total images	60,000
Training images	50,000
Test images	10,000
Image size	32 × 32 pixels
Color channels	3 (RGB)
Classes	10
Images per class	6,000 (5,000 train + 1,000 test)

The dataset contains the following 10 mutually exclusive object categories:

Airplane, Automobile, Bird, Cat, Deer, Dog, Frog, Horse, Ship,Truck

This mix intentionally includes:

• Animals (biological objects)

• Vehicles (man-made objects)

Step 1: Load Dataset

import tensorflow as tf
from tensorflow.keras.datasets import cifar10
import matplotlib.pyplot as plt

(x_train, y_train), (x_test, y_test) = cifar10.load_data()
print(x_train.shape, y_train.shape)

Downloading data from https://www.cs.toronto.edu/~kriz/cifar-10-python.tar.gz
170498071/170498071 ━━━━━━━━━━━━━━━━━━━━ 25s 0us/step
(50000, 32, 32, 3) (50000, 1)

Step 2: Visualize Sample Images (Human View)

Before training any model, it is critical to visually inspect the data. This helps confirm:

• Images are correctly loaded

• Labels match what humans perceive

• Data quality is acceptable

class_names = ['airplane','automobile','bird','cat','deer','dog','frog','horse','ship','truck']

plt.figure(figsize=(10,4))
for i in range(5):
    plt.subplot(1,5,i+1)
    plt.imshow(x_train[i])
    plt.title(class_names[y_train[i][0]])
    plt.axis('off')
plt.show()

Step 3: Data Preprocessing

x_train = x_train.astype('float32') / 255.0
x_test  = x_test.astype('float32') / 255.0

Step 4: CNN Model Design

from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Conv2D, MaxPooling2D, Flatten, Dense

model = Sequential([
 Conv2D(32,(3,3),activation='relu',input_shape=(32,32,3)),
 MaxPooling2D((2,2)),
 Conv2D(64,(3,3),activation='relu'),
 MaxPooling2D((2,2)),
 Flatten(),
 Dense(128,activation='relu'),
 Dense(10,activation='softmax')
])
model.summary()

C:\Users\Suyashi144893\AppData\Local\anaconda3\Lib\site-packages\keras\src\layers\convolutional\base_conv.py:113: UserWarning: Do not pass an `input_shape`/`input_dim` argument to a layer. When using Sequential models, prefer using an `Input(shape)` object as the first layer in the model instead.
  super().__init__(activity_regularizer=activity_regularizer, **kwargs)

Step 5: Compile and Train

model.compile(optimizer='adam',loss='sparse_categorical_crossentropy',metrics=['accuracy'])
model.fit(x_train,y_train,epochs=10,batch_size=64,validation_split=0.1)

Epoch 1/10
704/704 ━━━━━━━━━━━━━━━━━━━━ 9s 11ms/step - accuracy: 0.4713 - loss: 1.4763 - val_accuracy: 0.5736 - val_loss: 1.1962
Epoch 2/10
704/704 ━━━━━━━━━━━━━━━━━━━━ 8s 11ms/step - accuracy: 0.6038 - loss: 1.1313 - val_accuracy: 0.6352 - val_loss: 1.0437
Epoch 3/10
704/704 ━━━━━━━━━━━━━━━━━━━━ 8s 12ms/step - accuracy: 0.6544 - loss: 0.9925 - val_accuracy: 0.6668 - val_loss: 0.9771
Epoch 4/10
704/704 ━━━━━━━━━━━━━━━━━━━━ 8s 11ms/step - accuracy: 0.6880 - loss: 0.8975 - val_accuracy: 0.6932 - val_loss: 0.9041
Epoch 5/10
704/704 ━━━━━━━━━━━━━━━━━━━━ 8s 11ms/step - accuracy: 0.7138 - loss: 0.8240 - val_accuracy: 0.6942 - val_loss: 0.8921
Epoch 6/10
704/704 ━━━━━━━━━━━━━━━━━━━━ 8s 11ms/step - accuracy: 0.7383 - loss: 0.7479 - val_accuracy: 0.6978 - val_loss: 0.8793
Epoch 7/10
704/704 ━━━━━━━━━━━━━━━━━━━━ 8s 11ms/step - accuracy: 0.7616 - loss: 0.6823 - val_accuracy: 0.7048 - val_loss: 0.8695
Epoch 8/10
704/704 ━━━━━━━━━━━━━━━━━━━━ 8s 11ms/step - accuracy: 0.7854 - loss: 0.6196 - val_accuracy: 0.7034 - val_loss: 0.8874
Epoch 9/10
704/704 ━━━━━━━━━━━━━━━━━━━━ 8s 11ms/step - accuracy: 0.8022 - loss: 0.5649 - val_accuracy: 0.7094 - val_loss: 0.9191
Epoch 10/10
704/704 ━━━━━━━━━━━━━━━━━━━━ 8s 11ms/step - accuracy: 0.8258 - loss: 0.5070 - val_accuracy: 0.7112 - val_loss: 0.8951

<keras.src.callbacks.history.History at 0x193eedce790>

Step 6: Evaluation

loss,acc = model.evaluate(x_test,y_test)
print('Test Accuracy:',acc)

313/313 ━━━━━━━━━━━━━━━━━━━━ 1s 3ms/step - accuracy: 0.7027 - loss: 0.9355
Test Accuracy: 0.7027000188827515

Reflection

Why is visual inspection of data important before training a CNN?