This project demonstrates image classification using Convolutional Neural Networks (CNNs) with Keras and TensorFlow.
The goal is to classify traffic signs into 8 predefined categories using an ideal (noise-free) dataset.
For an introduction to CNNs, see: Understanding CNNs
- Dataset: Traffic sign images divided into training and testing sets.
- Model: A 6-layer CNN built using Keras.
- Training: Model trained and weights saved in
.h5format. - Prediction: Classifies new images and outputs results into a CSV file.
We use Keras ImageDataGenerator, which provides convenient data augmentation techniques such as resizing, rotation, flipping, and zooming to improve generalization.
📖 Documentation: ImageDataGenerator
import tensorflow as tf
from keras.models import Sequential
from keras.layers import Conv2D, MaxPooling2D, Flatten, Dense, Dropout
from keras.preprocessing.image import ImageDataGenerator
train_datagen = ImageDataGenerator(
rescale=1./255,
shear_range=0.2,
zoom_range=0.2,
horizontal_flip=True
)
test_datagen = ImageDataGenerator(rescale=1./255)
training_set = train_datagen.flow_from_directory(
'data/train',
target_size=(64, 64),
batch_size=32,
class_mode='categorical'
)
test_set = test_datagen.flow_from_directory(
'data/public_test',
target_size=(64, 64),
batch_size=32,
class_mode='categorical'
)2. Building the Model
Since this is a relatively simple classification task with clean data, the model is kept lightweight and efficient.
# Initialize CNN
classifier = Sequential()
# Convolution + Pooling Layers
classifier.add(Conv2D(32, (3, 3), input_shape=(64, 64, 3), activation='relu'))
classifier.add(MaxPooling2D(pool_size=(2, 2)))
classifier.add(Conv2D(32, (3, 3), activation='relu'))
classifier.add(MaxPooling2D(pool_size=(2, 2)))
classifier.add(Conv2D(64, (3, 3), activation='relu'))
classifier.add(MaxPooling2D(pool_size=(2, 2)))
# Flattening
classifier.add(Flatten())
# Fully Connected Layers
classifier.add(Dense(units=128, activation='relu'))
classifier.add(Dropout(0.5))
classifier.add(Dense(units=8, activation='softmax'))
# Compile the Model
classifier.compile(
optimizer='adam',
loss='categorical_crossentropy',
metrics=['accuracy']
)3. Training
The model is trained on the dataset, and weights are saved as my_model.h5.
classifier.fit_generator(
training_set,
steps_per_epoch=6589,
epochs=5,
validation_data=test_set,
validation_steps=20
)
classifier.save('my_model.h5')4. Image Prediction
New images located in data/data_private are classified using the trained model.
Predictions are saved in solve.csv in the format: <ImageID>,<Label>
import os, cv2, np
new_model = tf.keras.models.load_model('my_model.h5')
new_model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy'])
dirs = os.listdir('data/data_private')
with open("solve.csv", "a") as file:
for f in dirs:
file_name = os.path.join("data/data_private", f)
img = cv2.imread(file_name)
img = cv2.resize(img, (64, 64))
img = np.reshape(img, [1, 64, 64, 3])
classes = new_model.predict(img)
predicted_class = np.argmax(classes)
file.write(f"{f},{predicted_class}\n")⚡ Notes
- predict_classes() is deprecated in recent TensorFlow versions. Use np.argmax(model.predict(x)) instead.
- Data augmentation is crucial to improve generalization and avoid overfitting.
- For larger datasets, consider using transfer learning (e.g., VGG16, ResNet) for better performance.