def calculate_the_best_weight(self):
if os.listdir(checkpoint_path):
value = Image_Processing.extraction_image(checkpoint_path)
extract_num = [
os.path.splitext(os.path.split(i)[-1])[0] for i in value
]
num = [re.split('-', i) for i in extract_num]
accs = [float(i[-1]) for i in num]
losses = [float('-' + str(abs(float(i[-2])))) for i in num]
index = [acc + loss for acc, loss in zip(accs, losses)]
model_dict = dict((ind, val) for ind, val in zip(index, value))
return model_dict.get(max(index))
else:
logger.debug('没有可用的检查点')
from settings import MODEL_NAME
from settings import DATA_ENHANCEMENT
from settings import train_pack_path
from settings import validation_pack_path
from settings import test_pack_path
from settings import train_enhance_path
from Function_API import cheak_path
from Function_API import Image_Processing
from Function_API import parse_function_verification
gpus = tf.config.experimental.list_physical_devices(device_type='GPU')
for gpu in gpus:
tf.config.experimental.set_memory_growth(device=gpu, enable=True)
train_dataset = tf.data.TFRecordDataset(
Image_Processing.extraction_image(train_pack_path)).map(
parse_function_verification).batch(BATCH_SIZE)
validation_dataset = tf.data.TFRecordDataset(
Image_Processing.extraction_image(validation_pack_path)).map(
parse_function_verification).batch(BATCH_SIZE)
test_dataset = tf.data.TFRecordDataset(
Image_Processing.extraction_image(test_pack_path)).map(
parse_function_verification).batch(BATCH_SIZE)
model, c_callback = CallBack.callback(operator.methodcaller(MODEL)(Model))
model.summary()
model.compile(optimizer=tf.keras.optimizers.Adam(lr=1e-3, amsgrad=True),
from settings import train_path from Function_API import Image_Processing Image_Processing.rename_suffix(Image_Processing.extraction_image(train_path))
from settings import train_path from Function_API import Image_Processing Image_Processing.rename_path(train_path)
# 测试模型
import os
import random
import tensorflow as tf
from loguru import logger
from settings import BATCH_SIZE
from settings import model_path
from settings import test_pack_path
from settings import MODEL_LEAD_NAME
from settings import test_path
from Function_API import Image_Processing
from Function_API import Distinguish_image
from Function_API import parse_function_verification
gpus = tf.config.experimental.list_physical_devices(device_type='GPU')
for gpu in gpus:
tf.config.experimental.set_memory_growth(device=gpu, enable=True)
test_dataset = tf.data.TFRecordDataset(
Image_Processing.extraction_image(test_pack_path)).map(
parse_function_verification).batch(BATCH_SIZE)
model_path = os.path.join(model_path, MODEL_LEAD_NAME)
test_image_list = Image_Processing.extraction_image(test_path)
random.shuffle(test_image_list)
for i in test_image_list[:50]:
Distinguish_image.distinguish_image(model_path, i)
model = tf.keras.models.load_model(model_path)
logger.info(model.evaluate(test_dataset))
import random from settings import train_path from Function_API import Image_Processing train_image = Image_Processing.extraction_image(train_path) random.shuffle(train_image) Image_Processing.move_path(train_image)
from settings import validation_path
from settings import test_path
from settings import IMAGE_HEIGHT
from settings import IMAGE_WIDTH
from settings import train_enhance_path
from settings import DATA_ENHANCEMENT
from settings import TFRecord_train_path
from settings import TFRecord_validation_path
from settings import TFRecord_test_path
from Function_API import Image_Processing
from Function_API import WriteTFRecord
from concurrent.futures import ThreadPoolExecutor
if DATA_ENHANCEMENT:
with ThreadPoolExecutor(max_workers=100) as t:
for i in Image_Processing.extraction_image(train_path):
task = t.submit(Image_Processing.preprosess_save_images, i,
[IMAGE_HEIGHT, IMAGE_WIDTH])
train_image = Image_Processing.extraction_image(train_enhance_path)
random.shuffle(train_image)
train_lable = Image_Processing.extraction_one_hot_lable(train_image)
else:
train_image = Image_Processing.extraction_image(train_path)
random.shuffle(train_image)
train_lable = Image_Processing.extraction_one_hot_lable(train_image)
validation_image = Image_Processing.extraction_image(validation_path)
validation_lable = Image_Processing.extraction_one_hot_lable(validation_image)
test_image = Image_Processing.extraction_image(test_path)
test_lable = Image_Processing.extraction_one_hot_lable(test_image)