def check_predict(id='6100_3_2'):
model = get_unet()
model.load_weights('weights/unet_10_jk0.7914')
# 这个阈值是calc_jacc时算出来的最佳阈值
msk = predict_id(id, model, [0.4, 0.1, 0.4, 0.3, 0.3, 0.5, 0.3, 0.6, 0.1, 0.1])
img = M(id)
display_predict_result(img, msk)
def get_model(shape):
n_ch = shape[1]
patch_height = shape[2]
patch_width = shape[3]
if get_net == "unet":
model = network.get_unet(n_ch, patch_height, patch_width)
elif get_net == "fcn":
model = network.get_fcn(n_ch, patch_height, patch_width)
elif get_net == "hed":
model = network.get_hed(n_ch, patch_height, patch_width)
elif get_net == "unet3":
model = network.get_unet3(n_ch, patch_height, patch_width)
elif get_net == "unet4":
model = network.get_unet4(n_ch, patch_height, patch_width)
elif get_net == "unet5":
model = network.get_unet5(n_ch, patch_height, patch_width)
elif get_net == "unet_all":
model = network.get_unet_all(n_ch, patch_height, patch_width)
elif get_net == "unet_dm":
model = network.get_unet_dm(n_ch, patch_height, patch_width)
elif get_net == "unet_dm2":
model = network.get_unet_dm2(n_ch, patch_height, patch_width)
elif get_net == "unet_dsm":
model = network.get_unet_dsm(n_ch, patch_height, patch_width)
elif get_net == "unet_br":
model = network.get_unet_br(n_ch, patch_height, patch_width)
elif get_net == "unet_br2":
model = network.get_unet_br2(n_ch, patch_height, patch_width)
elif get_net == "fcnet":
model = network.get_fcnet(n_ch, patch_height, patch_width)
elif get_net == "unet_5l":
model = network.get_unet5l(n_ch, patch_height, patch_width)
elif get_net == "unet_6l":
model = network.get_unet6l(n_ch, patch_height, patch_width)
elif get_net == "unet_brnew":
model = network.get_unet_brnew(n_ch, patch_height, patch_width)
elif get_net == "hed_crf":
model = network.get_hedcrf(n_ch, patch_height, patch_width)
elif get_net == "unet3_com":
model = network.get_unet3_combine(n_ch, patch_height, patch_width)
elif get_net == "demo":
model = network.get_demo(n_ch, patch_height, patch_width)
else:
print("Please input a correct network!")
exit(0)
print("Check: final output of the network:", model.output_shape)
json_string = model.to_json()
open('./' + name_experiment + '/' + name_experiment + '_architecture.json',
'w').write(json_string)
return model
def train_net():
print("start train net")
x_val, y_val = np.load('data/x_tmp_%d.npy' % class_number), np.load('data/y_tmp_%d.npy' % class_number)
img, msk = np.load('data/x_trn_%d.npy' % class_number), np.load('data/y_trn_%d.npy' % class_number)
x_trn, y_trn = get_patches(img, msk)
model = get_unet()
model.load_weights('weights/unet_10_jk0.7878')
# 回调函数，在epoch结束后保存在验证集上性能最好的模型
model_checkpoint = ModelCheckpoint('weights/unet_tmp.hdf5', monitor='loss', save_best_only=True)
model.fit(x_trn, y_trn, batch_size=64, epochs=1, verbose=1, shuffle=True,
callbacks=[model_checkpoint], validation_data=(x_val, y_val))
score, trs = calc_jacc(model)
print('val jk', score)
model.save_weights('weights/unet_10_jk%.4f' % score)
return model
def check_predict(id='6120_2_3'):
model = get_unet()
model.load_weights('weights/unet_10_jk0.7878')
msk = predict_id(id, model, [0.4, 0.1, 0.4, 0.3, 0.3, 0.5, 0.3, 0.6, 0.1, 0.1])
img = M(id)
plt.figure()
ax1 = plt.subplot(131)
ax1.set_title('image ID:6120_2_3')
ax1.imshow(img[:, :, 5], cmap=plt.get_cmap('gist_ncar'))
ax2 = plt.subplot(132)
ax2.set_title('predict bldg pixels')
ax2.imshow(msk[0], cmap=plt.get_cmap('gray'))
ax3 = plt.subplot(133)
ax3.set_title('predict bldg polygones')
ax3.imshow(mask_for_polygons(mask_to_polygons(msk[0], epsilon=1), img.shape[:2]), cmap=plt.get_cmap('gray'))
plt.show()
def train_net():
print("start train net")
x_val, y_val = np.load('data/x_tmp_%d.npy' % N_Cls), np.load('data/y_tmp_%d.npy' % N_Cls)
img = np.load('data/x_trn_%d.npy' % N_Cls)
msk = np.load('data/y_trn_%d.npy' % N_Cls)
x_trn, y_trn = get_patches(img, msk)
model = get_unet()
model.load_weights('weights/unet_10_jk0.7878')
model_checkpoint = ModelCheckpoint('weights/unet_tmp.hdf5', monitor='loss', save_best_only=True)
for i in range(1):
model.fit(x_trn, y_trn, batch_size=64, nb_epoch=1, verbose=1, shuffle=True,
callbacks=[model_checkpoint], validation_data=(x_val, y_val))
del x_trn
del y_trn
x_trn, y_trn = get_patches(img, msk)
score, trs = calc_jacc(model)
print('val jk', score)
model.save_weights('weights/unet_10_jk%.4f' % score)
return model