import os
import sys
sys.path.append(os.path.join(os.path.dirname(__file__), '../'))
from drcn.graph import Graph
import tensorflow as tf
from drcn import args
from utils.load_data import load_all_data
import pickle
p_c_index, h_c_index, p_w_index, h_w_index, p_w_vec, h_w_vec, same_word, label = load_all_data(
'../input/train.csv', data_size=None)
p_c_index_evl, h_c_index_evl, p_w_index_evl, h_w_index_evl, p_w_vec_evl, h_w_vec_evl, same_word_evl, label_evl = load_all_data(
'../input/dev.csv', data_size=2000)
p_c_index_holder = tf.placeholder(name='p_c_index',
shape=(None, args.max_char_len),
dtype=tf.int32)
h_c_index_holder = tf.placeholder(name='h_c_index',
shape=(None, args.max_char_len),
dtype=tf.int32)
p_w_index_holder = tf.placeholder(name='p_w_index',
shape=(None, args.max_word_len),
dtype=tf.int32)
h_w_index_holder = tf.placeholder(name='h_w_index',
shape=(None, args.max_word_len),
dtype=tf.int32)
p_w_vec_holder = tf.placeholder(name='p_w_vec',
shape=(None, args.max_word_len,
args.word_embedding_len),
NUM_CLS = 2
parser = argparse.ArgumentParser()
parser.add_argument("--test_map_path", type=str)
parser.add_argument("--pred_path", type=str)
parser.add_argument("--model_path", type=str)
args = parser.parse_args()
TEST_MAP_PATH = args.test_map_path
PRED_PATH = args.pred_path
STRIDE = 20
model = vgg16()
model.load_weights(args.model_path)
test_data, img_names = load_data.load_all_data(MAP_PATH,
'',
WIN_SIZE,
20,
flip=False)
test_data = preprocess_input(test_data.astype('float64'))
pred = model.predict(test_data)
map_img = cv2.imread(MAP_PATH)
res = np.zeros((map_img.shape[0], map_img.shape[1]))
for i in range(pred.shape[0]):
if pred[i, 0] > 0.99:
idx = img_names[i].split('_')
x, y = int(idx[0]), int(idx[1])
res[x:x + WIN_SIZE, y:y + WIN_SIZE] = 255
cv2.imwrite(PRED_PATH, res)
import os
import sys
sys.path.append(os.path.join(os.path.dirname(__file__), '../'))
from drcn.graph import Graph
import tensorflow as tf
from drcn import args
from utils.load_data import load_all_data
import numpy as np
import pickle
os.environ['CUDA_VISIBLE_DEVICES'] = '1'
p_c_index, h_c_index, p_w_index, h_w_index, p_w_vec, h_w_vec, same_word, label = load_all_data('../input/test.csv',
data_size=None)
p_c_index_holder = tf.placeholder(name='p_c_index', shape=(None, args.max_char_len), dtype=tf.int32)
h_c_index_holder = tf.placeholder(name='h_c_index', shape=(None, args.max_char_len), dtype=tf.int32)
p_w_index_holder = tf.placeholder(name='p_w_index', shape=(None, args.max_word_len), dtype=tf.int32)
h_w_index_holder = tf.placeholder(name='h_w_index', shape=(None, args.max_word_len), dtype=tf.int32)
p_w_vec_holder = tf.placeholder(name='p_w_vec', shape=(None, args.max_word_len, args.word_embedding_len),
dtype=tf.float32)
h_w_vec_holder = tf.placeholder(name='h_w_vec', shape=(None, args.max_word_len, args.word_embedding_len),
dtype=tf.float32)
same_word_holder = tf.placeholder(name='same_word', shape=(None,), dtype=tf.int32)
label_holder = tf.placeholder(name='y', shape=(None,), dtype=tf.float32)
dataset = tf.data.Dataset.from_tensor_slices(
(p_c_index_holder, h_c_index_holder, p_w_index_holder, h_w_index_holder, p_w_vec_holder, h_w_vec_holder,
same_word_holder, label_holder))
dataset = dataset.batch(args.batch_size).repeat(args.epochs)
drcn_params['max_features'] = 1700
drcn_params['word_max_features'] = 7300
drcn_params['word_embed_size'] = 100
drcn_params['embed_size'] = 100
word_embedding_matrix = load_word_embed(drcn_params['word_max_features'],drcn_params['word_embed_size'])
char_embedding_matrix = load_char_embed(drcn_params['max_features'],drcn_params['embed_size'])
drcn_params['embedding_matrix'] = char_embedding_matrix
drcn_params['word_embedding_matrix'] = word_embedding_matrix
params = drcn_params
backend = DRCN(params)
if model_name == "drcn":
p_c_index, h_c_index, p_w_index, h_w_index, same_word, y = load_all_data('./input/train.csv',maxlen=params['input_shapes'][0][0])
x = [p_c_index, h_c_index, p_w_index, h_w_index]
y = keras.utils.to_categorical(y,num_classes=params['num_classes'])
p_c_index_evl, h_c_index_evl, p_w_index_evl, h_w_index_evl, same_word_evl, y_eval = load_all_data('./input/dev.csv',maxlen=params['input_shapes'][0][0])
x_eval = [p_c_index_evl, h_c_index_evl, p_w_index_evl, h_w_index_evl]
y_eval = keras.utils.to_categorical(y_eval,num_classes=params['num_classes'])
p_c_index_test, h_c_index_test, p_w_index_test, h_w_index_test, same_word_test, y_test = load_all_data('./input/test.csv',maxlen=params['input_shapes'][0][0])
x_test = [p_c_index_test, h_c_index_test, p_w_index_test, h_w_index_test]
y_test = keras.utils.to_categorical(y_test,num_classes=params['num_classes'])
else:
p, h, y = load_char_data('input/train.csv', data_size=None,maxlen=params['input_shapes'][0][0])
x = [p,h]
y = keras.utils.to_categorical(y,num_classes=params['num_classes'])
p_eval, h_eval, y_eval = load_char_data('input/dev.csv', data_size=None,maxlen=params['input_shapes'][0][0])
x_eval = [p_eval,h_eval]
y_eval = keras.utils.to_categorical(y_eval,num_classes=params['num_classes'])
MODEL_PATH = '' batch_size = 81 wetlands = [145, 147, 149, 162, 169, 178, 372, 385] others = [4, 29, 52, 96, 116, 297, 365, 376] latent_dim = 32 intermediate_dim = 1024 num_cls = 2 optimizer = Adam(lr=LEARNING_RATE) initializer = 'glorot_normal' #'random_uniform' # optimizer = SGD(lr=LEARNING_RATE) # optimizer = Adagrad(lr=LEARNING_RATE) #################################################################################### # load data x_u, _ = load_data.load_all_data(MAP_PATH, MASK_PATH, IMG_SIZE, STRIDE) # x_u, _ = load_data.load_wetland_samples(SUBSET_PATH) np.random.shuffle(x_u) x_l, target_name = load_data.load_wetland_samples(TARGET_SAMPLE_DIR) # x_l = x_u[39:40] # x_l_aug = data_augmentation.data_aug(x_l, SHIFT_LIST, ROTATION_ANGLE) x_l_aug = x_l # x_u = data_augmentation.data_aug(x_l, SHIFT_LIST, rotation) print(x_l_aug.shape) np.random.shuffle(x_l_aug) x_l = np.reshape(x_l, [-1, IMG_SIZE * IMG_SIZE * 3]) x_l_aug = np.reshape(x_l_aug, [-1, IMG_SIZE * IMG_SIZE * 3]) x_u = np.reshape(x_u, [-1, IMG_SIZE * IMG_SIZE * 3]) image_size = x_u.shape[1]
# training data generation
import cv2
import numpy as np
from utils import load_data, data_augmentation, helper
import os
MAP_PATH = './data/orcutt_2018/CA_Orcutt_20180828_TM_geo.png'
MASK_PATH = './data/orcutt_2018/wetlands.png'
SUBSET_PATH = './data/orcutt_2018/neg_samples'
IMG_SIZE = 40
STRIDE = 20
num_imgs = 1000
helper.create_folder(SUBSET_PATH)
mask = cv2.imread(MASK_PATH)
print(mask.shape)
all_data, _ = load_data.load_all_data(MAP_PATH, MASK_PATH, IMG_SIZE, STRIDE, flip=True)
print(all_data.shape)
np.random.shuffle(all_data)
# helper.remove_files(SUBSET_PATH)
for i in range(num_imgs):
cv2.imwrite(os.path.join(SUBSET_PATH,str(i)+'.png'), all_data[i].reshape((IMG_SIZE, IMG_SIZE, 3)))
print('done...')
from utils import load_data, data_augmentation
import numpy as np
import os
import cv2
os.environ["CUDA_VISIBLE_DEVICES"] = "0"
TARGET_SAMPLE_DIR = "./data/bray_2001/target_samples"
MAP_PATH = './data/bray_2001/CA_Bray_100414_2001_24000_geo.png'
MASK_PATH = './data/bray_2001/wetlands.png'
SHIFT_LIST = [-20, -15, -10, -5, 0, 5, 10, 15, 20]
ROTATION_ANGLE = []
for i in range(0, 360, 90):
ROTATION_ANGLE.append(i)
IMG_SIZE = 80
STRIDE = 30
x_u, img_names = load_data.load_all_data(MAP_PATH, MASK_PATH, IMG_SIZE, STRIDE)
x_l, target_name = load_data.load_wetland_samples(TARGET_SAMPLE_DIR)
x_l_aug = data_augmentation.data_aug(x_l, SHIFT_LIST, ROTATION_ANGLE)