def calib_input(iter):
images = []
line = open(calib_image_list).readlines()
#print(line)
for index in range(0, calib_batch_size):
curline = line[iter * calib_batch_size + index]
#print(curline)
calib_image_name = curline.strip()
# read image as rgb, returns numpy array (32,32, 3)
image = cv2.imread(calib_image_dir + calib_image_name)
##swap channels: from BGR to RGB
#B, G, R = cv2.split(image)
#image = cv2.merge([R, G, B])
img_array = img_to_array(image, data_format=None)
# scale the pixel values to range 0 to 1.0
image2 = cfg.Normalize(img_array)
# reshape numpy array to be (32,32,3)
image2 = image2.reshape((image2.shape[0], image2.shape[1], 3))
images.append(image2)
return {"conv2d_1_input": images}
def calib_input(iter):
assert (int(iter) <= int(tot_num_images / calib_batch_size)
), "number of iterations must be <=20"
images = []
#print(line)
for index in range(0, calib_batch_size):
#print(iter-1, index)
curline = line[(iter - 1) * calib_batch_size + index]
#print(curline)
calib_image_name = curline.strip()
# read image as rgb, returns numpy array (32,32, 3)
filename = os.path.join(calib_image_dir, calib_image_name)
image = cv2.imread(filename)
##swap channels: from BGR to RGB
#B, G, R = cv2.split(image)
#image = cv2.merge([R, G, B])
#img_array = img_to_array(image, data_format=None)
# scale the pixel values to range 0 to 1.0
image2 = cfg.Normalize(image)
# reshape numpy array to be (32,32,3)
image2 = image2.reshape((image2.shape[0], image2.shape[1], 3))
images.append(image2)
return {"conv2d_1_input": images}
y_valid = to_categorical(y_valid, 10)
# check settings #DB
assert True, (len(x_train) > cfg.NUM_TRAIN_IMAGES)
assert True, (len(x_test) >= (cfg.NUM_TRAIN_IMAGES + cfg.NUM_VAL_IMAGES))
assert True, (cfg.NUM_TRAIN_IMAGES == cfg.NUM_VAL_IMAGES)
#################################################################################
# pre-process the data
x_test = np.asarray(x_test)
x_train = np.asarray(x_train)
x_valid = np.asarray(x_valid)
#Normalize and convert from BGR to RGB
x_train = cfg.Normalize(x_train)
x_test = cfg.Normalize(x_test)
x_valid = cfg.Normalize(x_valid)
##################################################################################################
# construct the callback to save only the *best* model to disk
# based on the validation loss
fname = os.path.sep.join([weights, "best_chkpt.hdf5"])
checkpoint = ModelCheckpoint(
fname,
monitor="val_loss",
mode="min",
#monitor="val_acc", mode="max",
save_best_only=True,
verbose=1)
def graph_eval(input_graph_def, input_node, output_node):
#Reading image paths
test_img_paths = [
img_path for img_path in glob.glob(TEST_DIR + "/*/*.png")
]
NUMEL = len(test_img_paths)
assert (NUMEL > 0)
y_test = np.zeros((NUMEL, 1), dtype="uint8")
x_test = np.zeros((NUMEL, cfg.IMAGE_HEIGHT, cfg.IMAGE_WIDTH, 3),
dtype="uint8")
i = 0
for img_path in test_img_paths:
img = cv2.imread(img_path, cv2.IMREAD_COLOR)
##swap channels: from BGR to RGB
#B, G, R = cv2.split(img)
#img = cv2.merge([R, G, B])
img_array = img_to_array(img, data_format=None)
filename = os.path.basename(img_path)
class_name = filename.split("_")[0]
label = cfg.labelNames_dict[class_name]
#print("filename: ", img_path)
#print("classname: ", class_name)
x_test[i] = img_array
y_test[i] = int(label)
i = i + 1
x_test = cfg.Normalize(x_test)
#print(x_test[0])
#collect garbage to save memory #DB
#del img
#del test_img_paths
#del img_path
#gc.collect()
x_test = np.reshape(x_test, [-1, cfg.IMAGE_HEIGHT, cfg.IMAGE_WIDTH, 3])
y_test = tf.keras.utils.to_categorical(y_test, num_classes=cfg.NUM_CLASSES)
tf.compat.v1.import_graph_def(input_graph_def, name='')
# Get input placeholders & tensors
images_in = tf.compat.v1.get_default_graph().get_tensor_by_name(
input_node + ':0')
labels = tf.compat.v1.placeholder(tf.int32, shape=[None, cfg.NUM_CLASSES])
# get output tensors
logits = tf.compat.v1.get_default_graph().get_tensor_by_name(output_node +
':0')
# top 5 and top 1 accuracy
in_top5 = tf.nn.in_top_k(predictions=logits,
targets=tf.argmax(labels, 1),
k=5)
in_top1 = tf.nn.in_top_k(predictions=logits,
targets=tf.argmax(labels, 1),
k=1)
top5_acc = tf.reduce_mean(tf.cast(in_top5, tf.float32))
top1_acc = tf.reduce_mean(tf.cast(in_top1, tf.float32))
# Create the Computational graph
with tf.compat.v1.Session() as sess:
sess.run(tf.compat.v1.initializers.global_variables())
feed_dict = {images_in: x_test, labels: y_test}
t5_acc, t1_acc = sess.run([top5_acc, top1_acc], feed_dict)
#print(dir(x_test))
#print(max(x_test[0]))
#print(min(x_test[0]))
print(' Top 1 accuracy with validation set: {:1.4f}'.format(t1_acc))
print(' Top 5 accuracy with validation set: {:1.4f}'.format(t5_acc))
print('FINISHED!')
return
