def __getitem__(self, idx):
img_root = self.data_list[idx]
visit_array = visit_transform(img_root,self.args)
name = img_root.split('/')[-1]
label = int(img_root.split('/')[-2]) -1
img = cv2.imread(img_root)
if self.phase == 'train' and self.args.data_augmentation:
img = augmentation(img)
img = img / 255.0
img= np.transpose(img,(2,0,1))
return np.array(img), np.array(visit_array), label, name
def TakeOnePatch(image, label, patch_size):
"""
image,label: take patch from these image and label
patch_size: a int number, patch size is (patch_size,patch_size,patch_size) by default
return: corresponding image_patch and label_patch
"""
image_data = image.get_data()
label_data = label.get_data()
x, y, z = label_data.shape
i = np.random.randint(0, x - patch_size)
j = np.random.randint(0, y - patch_size)
k = np.random.randint(0, z - patch_size)
image_cube_data = image_data[i:i + patch_size, j:j + patch_size,
k:k + patch_size]
label_cube_data = label_data[i:i + patch_size, j:j + patch_size,
k:k + patch_size]
image_patch, label_patch = augmentation(image_cube_data, label_cube_data)
return image_patch, label_patch
import numpy as np
from models import Basic_Model, Hid_Model, Conv_Model, Conv_Model2, Conv_Model_WDDOBN, Conv_Model2_WDDOBN, Res_Model
from functions import *
from data_augmentation import augmentation
from fashion_mnist import load_fashion_mnist
import pickle
(x_train, y_train), (x_test, y_test) = load_fashion_mnist(normalize=True, flatten=False, one_hot_label=True)
x_train, y_train = augmentation(x_train, y_train, hflip=True, vflip=False)
model = Basic_Model()
#model = pickle.load(open("result/201905111739/model.pkl", 'rb'))
model.fit(x_train, y_train, x_test, y_test)
model.save_all()
from np import *
from config import GPU
from models import Basic_Model, Hid_Model, Conv_Model, Conv_Model2, Conv_Model32, Conv_Model_WDDOBN, Conv_Model2_WDDOBN, Res_Model, W_Res_Model
from functions import *
from fashion_mnist import load_fashion_mnist
import pickle
from data_augmentation import augmentation
(x_train, y_train), (x_test, y_test) = load_fashion_mnist(normalize=True,
flatten=False,
one_hot_label=True)
if GPU:
x_train, x_test = to_gpu(x_train), to_gpu(x_test)
y_train, y_test = to_gpu(y_train), to_gpu(y_test)
x_train, y_train = augmentation(x_train,
y_train,
hflip=False,
vflip=False,
hshift=True,
vshift=True)
model = Conv_Model32()
#model = pickle.load(open("result/201905111739/model.pkl", 'rb'))
model.fit(x_train, y_train, x_test, y_test)
model.save_all()
if original_img is None:
print('Input image error')
print(common_params.images_dir + '/train/rgb/' + fname + '.png')
sys.exit(1)
# 画像をSSDの入力サイズにリサイズ
original_img = cv.resize(original_img, (common_params.insize, common_params.insize), interpolation = cv.INTER_CUBIC)
for ag in range(0, common_params.augmentation_factor):
input_img = original_img.copy()
idx = original_idx #クラス
gt_boxes = original_gt_boxes
if (ag >= 1):
input_img, idx, gt_boxes, border_pixels, crop_param, hsv_param, flip_type = augmentation(input_img, idx, gt_boxes)
if visible:
out2 = input_img.copy()
for bx in range(0, len(gt_boxes)):
p1 = int(gt_boxes[bx][0])
p2 = int(gt_boxes[bx][1])
p3 = int(gt_boxes[bx][2])
p4 = int(gt_boxes[bx][3])
cv.rectangle(out2, (p1, p2), (p3, p4), (0, 255, 0), 2)
cv.imshow('Augmentation', out2)
cv.waitKey(10)
img_width = input_img.shape[1]
img_height = input_img.shape[0]