def black_test(self):
dataprovider = imageUnits.ImageProvider(path='data/val/*.tif', bathsize=50, shuffle_data=False, channels=2,
nclass=3, pad=False)
sum_acc = 0
for i in range(16):
batch_x, batch_y = dataprovider.next_batch()
zero = np.zeros((50, 64, 64, 1))
one = np.ones((50, 64, 64, 1))
batch_x = np.concatenate((zero, one), axis=3)
accuracy = np.mean(
np.equal(np.argmax(batch_x, 3), np.argmax(batch_y, 3)).astype(np.float))
sum_acc += accuracy
print(sum_acc / 16)
import FCN
import imageUnits
batch_size = 1
channels = 3
nclass = 2
path = 'data/train/*.tif'
data_provider = imageUnits.ImageProvider(path=path,
bathsize=batch_size,
shuffle_data=True,
channels=channels,
nclass=nclass)
net = FCN.FC_net(data_provider,
batch_size,
conv_size=3,
pool_size=2,
channels=channels,
nclass=nclass,
save_path='model',
white_channel_weight=0.5,
layers=5)
net.predite()
import imageUnits
import tensorflow as tf
import numpy as tf
import Unet
train_path = 'data/val/*.tif'
batch_size = 1
channels = 3
nclass = 2
pad = True
predition_data_provider = imageUnits.ImageProvider(path=train_path,
bathsize=batch_size,
shuffle_data=True,
channels=channels,
nclass=nclass,
start=0,
pad=pad)
unet = Unet.unet(predition_data_provider,
layers=3,
first_feature_num=64,
conv_size=3,
pool_size=2,
batch_size=batch_size,
channels=channels,
nclass=nclass,
save_path='layer3',
white_channel_weight=0.4,
pad=pad)
unet.predite()