77 lines
3.0 KiB
Plaintext
77 lines
3.0 KiB
Plaintext
|
|
||
|
|
[1]. Tool: Claude
|
||
|
|
Prompt: How can I efficiently import a large amount of JPEG images into python for use with tensorflow while normalizing and randomizing the dataset?
|
||
|
|
Output:
|
||
|
|
from tensorflow.keras.preprocessing import image_dataset_from_directory
|
||
|
|
|
||
|
|
# Load color JPEG images from directory structure:
|
||
|
|
# Data/
|
||
|
|
# ├── damage/
|
||
|
|
# └── no_damage/
|
||
|
|
|
||
|
|
dataset = image_dataset_from_directory(
|
||
|
|
'Data',
|
||
|
|
image_size=(224, 224),
|
||
|
|
batch_size=32,
|
||
|
|
label_mode='binary',
|
||
|
|
color_mode='rgb', # Explicitly load as color (this is the default)
|
||
|
|
shuffle=True,
|
||
|
|
seed=42
|
||
|
|
)
|
||
|
|
|
||
|
|
# Normalize pixel values to [0, 1] range for better training
|
||
|
|
normalization_layer = tf.keras.layers.Rescaling(1./255)
|
||
|
|
dataset = dataset.map(lambda x, y: (normalization_layer(x), y))
|
||
|
|
|
||
|
|
# Optimize performance
|
||
|
|
dataset = dataset.prefetch(tf.data.AUTOTUNE)
|
||
|
|
|
||
|
|
print(f"Dataset created. Each batch shape: (batch_size, 224, 224, 3)")
|
||
|
|
|
||
|
|
[2]. Tool: Google AI Mode
|
||
|
|
Prompt: How can I ensure my GPU memory is emptied between model training sessions?
|
||
|
|
Output: import tensorflow as tf
|
||
|
|
import gc
|
||
|
|
|
||
|
|
# Load and use your model
|
||
|
|
# ...
|
||
|
|
|
||
|
|
del model # Delete the model object
|
||
|
|
tf.keras.backend.clear_session() # Clear the Keras backend session
|
||
|
|
gc.collect() # Force garbage collection
|
||
|
|
|
||
|
|
[2]. Tool: Google AI Mode
|
||
|
|
Prompt: vgg16 with binary classification
|
||
|
|
Output: import tensorflow as tf
|
||
|
|
from tensorflow.keras.applications import VGG16
|
||
|
|
from tensorflow.keras.models import Sequential, Model
|
||
|
|
from tensorflow.keras.layers import Dense, Flatten, Dropout
|
||
|
|
from tensorflow.keras.optimizers import Adam
|
||
|
|
from tensorflow.keras.losses import binary_crossentropy
|
||
|
|
|
||
|
|
# 1. Load VGG16 without the top layer
|
||
|
|
IMG_SIZE = (224, 224) # VGG16 default input size
|
||
|
|
vgg_base = VGG16(weights='imagenet', include_top=False, input_shape=IMG_SIZE + (3,))
|
||
|
|
|
||
|
|
# 2. Freeze the convolutional base
|
||
|
|
vgg_base.trainable = False
|
||
|
|
|
||
|
|
# 3. Create a new model with custom top layers
|
||
|
|
model = Sequential()
|
||
|
|
model.add(vgg_base)
|
||
|
|
model.add(Flatten())
|
||
|
|
model.add(Dense(256, activation='relu'))
|
||
|
|
model.add(Dropout(0.5)) # Regularization to prevent overfitting
|
||
|
|
model.add(Dense(1, activation='sigmoid')) # Single neuron with sigmoid for binary classification
|
||
|
|
|
||
|
|
# 4. Compile the model
|
||
|
|
model.compile(
|
||
|
|
optimizer=Adam(learning_rate=0.0001), # Lower learning rate often better for transfer learning
|
||
|
|
loss=binary_crossentropy,
|
||
|
|
metrics=['accuracy']
|
||
|
|
)
|
||
|
|
|
||
|
|
# 5. Train the model (using ImageDataGenerator for data preparation)
|
||
|
|
# ... data loading and augmentation code here ...
|
||
|
|
# model.fit(...)
|