def predict_class(img_path):
# prepare image for classification using keras utility functions
image = load_img(img_path, target_size=target_size)
image_arr = img_to_array(image) # convert from PIL Image to NumPy array
image_arr /= 255
# to be able to pass it through the network and use batches, we want it with shape (1, 224, 224, 3)
image_arr = np.expand_dims(image_arr, axis=0)
# print(image.shape)
# build the VGG16 network
model = applications.VGG16(include_top=False, weights='imagenet')
# get the bottleneck prediction from the pre-trained VGG16 model
bottleneck_features = model.predict(image_arr)
# build top model
model = create_top_model("softmax", bottleneck_features.shape[1:])
model.load_weights("res/_top_model_weights.h5")
predicted = model.predict(bottleneck_features)
# predicted_onehot = to_categorical(predicted, num_classes=num_classes)
return np.asarray(predicted[0]) # float32
batch_size=batch_size,
class_mode='categorical',
shuffle=False) # data is ordered
num_val_samples = len(val_generator.filenames)
# load vgg features
val_data = np.load('res/vgg_val_features.npy')
val_labels = val_generator.classes
val_labels_onehot = to_categorical(val_labels, num_classes=num_classes)
# ---------- CREATE AND TRAIN MODEL ----------
# create the top model to be trained
model = create_top_model("softmax", train_data.shape[1:])
model.compile(optimizer="rmsprop", loss="categorical_crossentropy", metrics=["accuracy"])
# only save the best weights. if the accuracy doesnt improve in 2 epochs, stop.
checkpoint_callback = ModelCheckpoint(
"res/top_model_weights.h5", # store weights with this file name
monitor='val_accuracy',
verbose=1,
save_best_only=True,
mode='max')
early_stop_callback = EarlyStopping(
monitor='val_accuracy',
patience=2, # max number of epochs to wait
mode='max')
'../../../dataset/split_data/test/',
target_size=target_size,
batch_size=batch_size,
class_mode='categorical', # specify categorical
shuffle=False) # data is ordered
# load vgg features
test_data = np.load('res/vgg_test_features.npy')
test_labels = test_generator.classes # actual class number
test_labels_onehot = to_categorical(
test_labels, num_classes=num_classes) # class number in onehot
# ---------- TEST MODEL ----------
model = create_top_model("softmax", test_data.shape[1:])
model.load_weights("res/_top_model_weights.h5")
model.compile(optimizer="rmsprop",
loss="categorical_crossentropy",
metrics=["accuracy"])
predicted = model.predict_classes(test_data)
# ---------- DISPLAY INFORMATION DEPENDING ON COMMAND LINE FLAGS ----------
if args.acc:
loss, acc = model.evaluate(test_data,
test_labels_onehot,
batch_size=batch_size,
verbose=1)
from keras.preprocessing.image import load_img, img_to_array
from helper import create_top_model, class_labels, target_size, num_classes
import numpy as np
from keras import applications
import operator
import matplotlib.pyplot as plt
from keras.utils.np_utils import to_categorical
import os
# iterate through all testing images
data_path = 'imgs/test'
csv_header = ['img','c0','c1','c2','c3','c4','c5','c6','c7','c8','c9']
all_entries = []
model = applications.VGG16(include_top=False, weights='imagenet')
model_top = create_top_model("softmax", (7,7,512))
model_top.load_weights("res/top_model_weights_final.h5")
def predict_class(img_path):
image = load_img(img_path, target_size=target_size)
image_arr = img_to_array(image) # convert from PIL Image to NumPy array
image_arr /= 255
image_arr = np.expand_dims(image_arr, axis=0)
predicted = model_top.predict(bottleneck_features)
# predicted_onehot = to_categorical(predicted, num_classes=num_classes)
return np.asarray(predicted[0]) # float32
for i, file in enumerate(os.listdir(data_path)):
file_name = os.fsdecode(file)
image_arr = img_to_array(
image) # convert from PIL Image to NumPy array
image_arr /= 255
# to be able to pass it through the network and use batches, we want it with shape (1, 224, 224, 3)
image_arr = np.expand_dims(image_arr, axis=0)
# print(image.shape)
# build the VGG16 network
model = applications.VGG16(include_top=False, weights='imagenet')
# get the bottleneck prediction from the pre-trained VGG16 model
bottleneck_features = model.predict(image_arr)
# build top model
model = create_top_model("softmax", bottleneck_features.shape[1:])
model.load_weights("res/top_model_weights.h5")
predicted = model.predict(bottleneck_features)
decoded_predictions = dict(zip(class_labels, predicted[0]))
decoded_predictions = sorted(decoded_predictions.items(),
key=operator.itemgetter(1),
reverse=True)
print()
x_coor = 5
y_coor = 10
count = 1
for key, value in decoded_predictions[:5]: