def generate_contrast_plus_test(path, base_path):
loadfile = load.LoadData('D:\\pythonProject\\image_data_map.csv')
pf, dict_index = loadfile.get_pixel_data()
image_path_set = os.listdir(path)
for label in image_path_set:
image_name_set = os.listdir(osp.join(path, label))
num = random.randint(400, 700)
image_name1 = image_name_set[0]
temp_label_set = list(image_name_set)
temp_label_set.remove(image_name1)
for i in range(num):
image_name2 = temp_label_set[random.randint(
0,
len(temp_label_set) - 1)]
index_1 = dict_index.get(image_name1)
index_2 = dict_index.get(image_name2)
print('index_1: ' + str(index_1) + ', index_2: ' + str(index_2))
if index_1 is None or index_2 is None:
continue
x_1, f_1 = load.get_pixel_by_index(index_1, pf)
x_2, f_2 = load.get_pixel_by_index(index_2, pf)
pre1 = image_name1.replace('.png', '')
pre2 = image_name2.replace('.png', '')
save_path = osp.join(base_path,
pre1 + '~' + pre2 + '~' + str(i) + 'tp.png')
generate_origin.process_contrast(x_1, f_1, x_2, f_2, save_path)
def generate_contrast_plus_train(img_path, base_path):
image_path_set = os.listdir(img_path)
print(img_path + ' has ' + str(len(image_path_set)) + ' labels ')
loadfile = load.LoadData('D:\\pythonProject\\image_data_map.csv')
pf, dict_index = loadfile.get_pixel_data()
for label in image_path_set:
image_name_set = os.listdir(osp.join(img_path, label))
label_num = len(image_name_set)
# choose first image as standard image todo standard image is not static
standard_image = image_name_set[0]
print(label + ' has ' + str(label_num) +
' images, standard image is ' + standard_image)
index_standard = dict_index.get(standard_image)
temp_image_set = list(image_name_set)
temp_image_set.remove(standard_image)
for image_name in temp_image_set:
index_image = dict_index.get(image_name)
print('index_1: ' + str(index_standard) + ', index_2: ' +
str(index_image))
if index_standard is None or index_image is None:
continue
print(label + ': generate ' + image_name + ' with ' +
standard_image)
x_1, f_1 = load.get_pixel_by_index(index_standard, pf)
x_2, f_2 = load.get_pixel_by_index(index_image, pf)
pre = image_name.replace('.png', '')
save_path = osp.join(base_path, pre + 'tp.png')
generate_origin.process_contrast(x_1, f_1, x_2, f_2, save_path)
def generate_contrast_minus_train(img_path, base_path):
image_path_set = os.listdir(img_path)
label_cnt = len(image_path_set)
print(img_path + ' has ' + str(label_cnt) + ' labels ')
loadfile = load.LoadData('D:\\pythonProject\\image_data_map.csv')
pf, dict_index = loadfile.get_pixel_data()
for label in image_path_set:
temp_label_set = list(image_path_set)
temp_label_set.remove(label)
image_name_set = os.listdir(osp.join(img_path, label))
for image_name in image_name_set:
minus_label = temp_label_set[random.randint(
0,
len(temp_label_set) - 1)]
minus_name_set = os.listdir(img_path + '\\' + minus_label)
minus_name = minus_name_set[random.randint(0,
len(minus_name_set) -
1)]
index_1 = dict_index.get(image_name)
index_2 = dict_index.get(minus_name)
print('index_1: ' + str(index_1) + ', index_2: ' + str(index_2))
if index_1 is None or index_2 is None:
continue
print(label + ': generate ' + image_name + ' with ' + minus_name)
x_1, f_1 = load.get_pixel_by_index(index_1, pf)
x_2, f_2 = load.get_pixel_by_index(index_2, pf)
pre = image_name.replace('.png', '')
save_path = osp.join(base_path, pre + 'tm.png')
generate_origin.process_contrast(x_1, f_1, x_2, f_2, save_path)
def draw_in_one(temp, pf, save_path):
img = 255 * np.ones((256, 256, 3), np.uint8)
for index in temp:
x, f = load.get_pixel_by_index(index, pf)
generate_origin.draw_line(img, x, f, (255, 0, 0))
# print(index)
img2 = cv.flip(img, 0, dst=None)
cv.imwrite(save_path, img2)
def generate_single_contrast_by_name(image_name1, image_name2, base_path):
loadfile = load.LoadData('D:\\pythonProject\\image_data_map.csv')
pf, dict_index = loadfile.get_pixel_data()
# file = open('D:\\graduationproject\\ver3\\similarity\\cbir\\error.txt')
# for line in file:
# content = line.split(' ')
# image_name1 = content[0]
# image_name2 = content[1]
# label = content[2].replace('\n', '')
index_1 = dict_index.get(image_name1)
index_2 = dict_index.get(image_name2)
print(str(index_1) + ';' + str(index_2))
# if index_1 is None or index_2 is None:
# continue
x_1, f_1 = load.get_pixel_by_index(index_1, pf)
x_2, f_2 = load.get_pixel_by_index(index_2, pf)
save_path = osp.join(base_path, image_name1 + '~' + image_name2 + '.png')
generate_origin.process_contrast(x_2, f_2, x_1, f_1, save_path)
def generate_contrast_minus_test(image_name, path, base_path):
loadfile = load.LoadData('D:\\pythonProject\\image_data_map.csv')
pf, dict_index = loadfile.get_pixel_data()
image_path_set = os.listdir(path)
num = random.randint(800, 1000)
for i in range(num):
minus_label = image_path_set[random.randint(0,
len(image_path_set) - 1)]
minus_name_set = os.listdir(osp.join(path, minus_label))
minus_name = minus_name_set[random.randint(0, len(minus_name_set) - 1)]
index_1 = dict_index.get(image_name)
index_2 = dict_index.get(minus_name)
print('index_1: ' + str(index_1) + ', index_2: ' + str(index_2))
if index_1 is None or index_2 is None:
continue
x_1, f_1 = load.get_pixel_by_index(index_1, pf)
x_2, f_2 = load.get_pixel_by_index(index_2, pf)
pre1 = image_name.replace('.png', '')
pre2 = minus_name.replace('.png', '')
save_path = osp.join(base_path,
pre1 + '~' + pre2 + '~' + str(i) + 'tm.png')
generate_origin.process_contrast(x_1, f_1, x_2, f_2, save_path)
def generate_triplet_origin_image(save_path, source_path):
loadfile = load.LoadData('D:\\pythonProject\\image_data_map.csv')
pf, dict_index = loadfile.get_pixel_data()
image_path_set = os.listdir(source_path)
image_set = set()
for image in image_path_set:
temp = image.split('~')
image_set.add(temp[0] + '.png')
image_set.add(temp[1] + '.png')
print('origin image num: ' + str(len(image_set)))
for image_name in image_set:
index = dict_index.get(image_name)
if index is None:
continue
x, f = load.get_pixel_by_index(index, pf)
generate_origin.process(x, f, osp.join(save_path, image_name))
print('origin image: ' + image_name)
def data_augmentation_from_map(image_name_set, dict_index, pf, path):
# order of magnitude
label_num = len(image_name_set)
num = int((4500 + random.randint(-700, 800)) / label_num)
# get data source
for image_name in image_name_set:
index_image = dict_index.get(image_name)
if index_image is None:
continue
x, f = load.get_pixel_by_index(index_image, pf)
pre = image_name.replace('.png', '')
for k in range(num):
length = len(f)
f_add = np.empty([length, 1], dtype=int)
for i in range(length):
temp = int(f[i] + random.randint(0, 3))
f_add[i] = temp if random.randint(
0, 1) <= 0.45 and temp <= max(f) else f[i]
add_path = osp.join(path, pre + '-' + str(k) + '.png')
process(x, f_add, add_path)