args = parser.parse_args()
train_data = myDataGenerator(args.timgs,
args.tmasks,
batch=args.batch,
imgsize=args.size)
numcls = train_data.getClsnum()
valid_data = myDataGenerator(args.vimgs,
args.vmasks,
numcls,
batch=args.batch,
imgsize=args.size)
model = myUnet(numcls, args.filters, args.droprate, args.convkernel,
args.transkernel, args.convstride, args.transstride,
args.convpadding, args.transpadding, args.activation,
args.batchnorm)
if args.checkpoint != '':
try:
model.load(args.checkpoint)
except Exception as e:
print(e)
exit()
lossobj = CategoricalCrossentropy()
optimizer = None
if args.optimizer.lower() == 'adadelta':
optimizer = Adadelta(args.learningrate)
from tqdm import tqdm
from sklearn.preprocessing import OneHotEncoder
from model import myUnet
from datapro import loaddata
from evaluation import Jaccard_eval
from repress import watershed_process, kluster_proess
from datapro import loadpro_img
result_path = 'dataset1/test_RES' #存储路径
model_path = 'unet1params3.pkl'
test_path = 'dataset1/test/'
repressmethod = 'connectedComponents' #选择后处理方法'watershed','cluster'
imgsize = 628
#加载模型
model = myUnet().cuda()
model.load_state_dict(torch.load(model_path))
test_list = sorted(
[os.path.join(test_path, img) for img in os.listdir(test_path)])
test_data = loadpro_img(test_list)
test_datapro = np.zeros(
(len(test_data), 1, imgsize + 92 * 2, imgsize + 92 * 2))
for i in range(len(test_data)):
test_datapro[i] = np.pad(test_data[i], ((92, 92)), 'symmetric')
test_datapro[i] = test_datapro[i] / 255.0
test_datapro[i] = (test_datapro[i] - 0.5) / 0.5
test_datapro = torch.from_numpy(test_datapro)
model.eval()
pred_all = []
i = 0
from keras.preprocessing.image import ImageDataGenerator, array_to_img, img_to_array, load_img
from PIL import Image
from model import myUnet
import cv2
import glob
import data
import math
import matplotlib.image as mpimg
Train_DIR = "E:\\Experiment data\\data\\Landsat\\Train\\"
Val_DIR = "E:\\Experiment data\\data\\Landsat\\Val\\"
Test_DIR = "E:\\Experiment data\\data\\Net3\\Test\\masked\\"
Result_DIR = "E:\\Experiment data\\data\\Net3\\Test\\resultMSCN\\"
model = myUnet()
model.load(r"G:\Detection\Net1\MSCN\weight\1000_weights_2018-11-16-17-16-07.h5")
imgsname = glob.glob(Test_DIR + "whole\\*.tif")
imgdatas = np.ndarray((1, 512, 512, 3), dtype=np.float32)
num = 0
patch_size = 512
def max_img(img):
x = np.zeros((patch_size, patch_size, 3))
# cloud = np.ndarray((patch_size, patch_size))
# shadow = np.ndarray((patch_size, patch_size))
# background = np.ndarray((patch_size, patch_size))
for i in range(0, patch_size):
for j in range(0, patch_size):