def generate_mean_pixel_file():
C = Config()
all_imgs, _, _ = get_data(ROI_BBOX_FILE)
avg = [0, 0, 0]
for img_data in all_imgs:
print(img_data['filepath'])
img_data_aug, x_img = augment(img_data, C, augment=False)
(width, height) = (img_data_aug['width'], img_data_aug['height'])
(rows, cols, _) = x_img.shape
# get image dimensions for resizing
(resized_width,
resized_height) = get_new_img_size(width, height, C.im_size)
# resize the image so that smalles side is length = 600px
x_img = cv2.resize(x_img, (resized_width, resized_height),
interpolation=cv2.INTER_CUBIC)
pixels = (resized_width * resized_height)
avg[0] += np.sum(x_img[:, :, 0]) / pixels
avg[1] += np.sum(x_img[:, :, 1]) / pixels
avg[2] += np.sum(x_img[:, :, 2]) / pixels
avg = [a / len(all_imgs) for a in list(avg)]
np.savetxt(MEAN_PIXEL_FILE, avg, delimiter=',')
def get_anchor_gt(all_img_data, class_count, C, backend, mode='train'): all_img_data = sorted(all_img_data) sample_selector = SampleSelector(class_count) while True: if mode == 'train': random.shuffle(all_img_data) for img_data in all_img_data: try: if C.balanced_classes and sample_selector.skip_sample_for_balanced_class(img_data): continue # read in image, and optionally add augmentation if mode == 'train': img_data_aug, x_img = data_augment.augment(img_data, C, augment=True) else: img_data_aug, x_img = data_augment.augment(img_data, C, augment=False) (width, height) = (img_data_aug['width'], img_data_aug['height']) (rows, cols, _) = x_img.shape assert cols == width assert rows == height # get image dimensions for resizing (resized_width, resized_height) = get_new_img_size(width, height, C.im_size) # resize the image so that smalles side is length = 600px x_img = cv2.resize(x_img, (resized_width, resized_height), interpolation=cv2.INTER_CUBIC) try: y_rpn_cls, y_rpn_regr = calc_rpn(C, img_data_aug, width, height, resized_width, resized_height) except: continue # Zero-center by mean pixel, and preprocess image x_img = x_img[:,:, (2, 1, 0)] # BGR -> RGB x_img = x_img.astype(np.float32) x_img[:, :, 0] -= C.img_channel_mean[0] x_img[:, :, 1] -= C.img_channel_mean[1] x_img[:, :, 2] -= C.img_channel_mean[2] x_img /= C.img_scaling_factor x_img = np.transpose(x_img, (2, 0, 1)) x_img = np.expand_dims(x_img, axis=0) y_rpn_regr[:, y_rpn_regr.shape[1]/2:, :, :] *= C.std_scaling if backend == 'tf': x_img = np.transpose(x_img, (0, 2, 3, 1)) y_rpn_cls = np.transpose(y_rpn_cls, (0, 2, 3, 1)) y_rpn_regr = np.transpose(y_rpn_regr, (0, 2, 3, 1)) yield np.copy(x_img), [np.copy(y_rpn_cls), np.copy(y_rpn_regr)], img_data_aug except Exception as e: print(e) continue
def get_anchor_gt(all_img_data,
class_mapping,
class_count,
C,
backend,
mode='train'):
downscale = float(C.rpn_stride)
all_img_data = sorted(all_img_data, key=sorted)
anchor_sizes = C.anchor_box_scales
anchor_ratios = C.anchor_box_ratios
num_anchors = len(anchor_sizes) * len(anchor_ratios)
sample_selector = SampleSelector(class_count)
while True:
if mode == 'train':
random.shuffle(all_img_data)
for img_data in all_img_data:
if C.balanced_classes and sample_selector.skip_sample_for_balanced_class(
img_data):
continue
# read in image, and optionally add augmentation
# img_data_aug, x_img = data_augment.augment(img_data, C, augment=True)
if mode == 'train':
img_data_aug, x_img = data_augment.augment(img_data,
C,
augment=True)
else:
img_data_aug, x_img = data_augment.augment(img_data,
C,
augment=False)
(width, height) = (img_data_aug['width'], img_data_aug['height'])
(rows, cols, _) = x_img.shape
assert cols == width
assert rows == height
# get image dimensions for resizing
(resized_width,
resized_height) = get_new_img_size(width, height, C.im_size)
# resize the image so that smalles side is length = 600px
x_img = cv2.resize(x_img, (resized_width, resized_height),
interpolation=cv2.INTER_CUBIC)
# calculate the output map size based on the network architecture
(output_width,
output_height) = get_img_output_length(resized_width,
resized_height)
try:
x_rois, y_rpn_cls, y_rpn_regr, y_class_num, y_class_regr = calcY(
C, class_mapping, img_data_aug, width, height,
resized_width, resized_height)
except:
continue
# Zero-center by mean pixel
x_img = x_img.astype(np.float32)
x_img[:, :, 0] -= 103.939
x_img[:, :, 1] -= 116.779
x_img[:, :, 2] -= 123.68
x_img = np.transpose(x_img, (2, 0, 1))
x_img = np.expand_dims(x_img, axis=0)
y_rpn_regr[:, int(y_rpn_regr.shape[1] / 2):, :, :] *= C.std_scaling
y_class_regr[:,
int(y_class_regr.shape[1] / 2):, :] *= C.std_scaling
if backend == 'tf':
x_img = np.transpose(x_img, (0, 2, 3, 1))
y_rpn_cls = np.transpose(y_rpn_cls, (0, 2, 3, 1))
y_rpn_regr = np.transpose(y_rpn_regr, (0, 2, 3, 1))
yield [np.copy(x_img), np.copy(x_rois)], [
np.copy(y_rpn_cls),
np.copy(y_rpn_regr),
np.copy(y_class_num),
np.copy(y_class_regr)
]
all_imgs = []
classes = {}
bbox_threshold = 0.8
visualise = True
#for idx, img_name in enumerate(sorted(os.listdir(img_path))):
# if not img_name.lower().endswith(('.bmp', '.jpeg', '.jpg', '.png', '.tif', '.tiff')):
# continue
for idx in range(64):
st = time.time()
_, img = data_augment.augment(n=np.random.randint(low=1, high=5))
X, ratio = format_img(img, C)
X = np.transpose(X, (0, 2, 3, 1))
img = (img * 255).astype(np.uint8)
img = cv2.cvtColor(img, cv2.COLOR_GRAY2RGB)
[Y1, Y2, F] = model_rpn.predict(X)
R = roi_helpers.rpn_to_roi(Y1,
Y2,
C,
K.image_dim_ordering(),
overlap_thresh=0.7)
R[:, 2] -= R[:, 0]
R[:, 3] -= R[:, 1]
def get_anchor_gt(all_img_data, class_count, C, img_length_calc_function, mode='train'):
#original code doen not compatible with py3.5
#all_img_data = sorted(all_img_data)
sample_selector = SampleSelector(class_count)
if mode == 'train':
#randomlized the data
np.random.shuffle(all_img_data)
#print("start")
for img_data in all_img_data:
try:
## C.balanced_classes = False for dafault
## bool for skip_
## some kinds of threshold
if C.balanced_classes and sample_selector.skip_sample_for_balanced_class(img_data):
continue
#read img and augment
## if train: hor/ver flip or rotate according to setting
## else: just read img and return img and its size
## img_data_aug 是哈希表，存放各种和当前图象有关的信息
## img 是cv2读进来的图片
#print("read")
if mode == 'train' :
img_data_aug, x_img = data_augment.augment(img_data, C, augment = True)
else:
img_data_aug, x_img = data_augment.augment(img_data, C, augment = False)
## check the information on side
(width, height) = (img_data_aug['width'], img_data_aug['height'])
(rows,cols, _ ) = x_img.shape
assert cols == width
assert rows == height
# get image dim
# C.im_size = 600
# resize and make the shorter side fixing at 600
(resized_w, resized_h) = get_new_img_size(width, height,C.im_size)
x_img = cv2.resize(x_img, (resized_w, resized_h), interpolation = cv2.INTER_CUBIC)
try:
y_rpn_cls, y_rpn_reg = calc_rpn(C, img_data_aug, width, height, resized_w, resized_h, img_length_calc_function)
except:
continue
############################################### 4.1 ################################################################
## Zero-mean preprocess
#BGR -> RGB due to cv2
x_img = x_img[:,:,(2,1,0)]
x_img = x_img.astype(np.float32)
#C.img_channel_mean should calculated by ourself
## 减去均值达成均值归一
# x_img[:,:,0] -= C.img_chanel_mean[0]
# x_img[:,:,1] -= C.img_chanel_mean[1]
# x_img[:,:,2] -= C.img_chanel_mean[2]
x_img[:,:,0] -= np.mean(x_img[:,:,0])
x_img[:,:,1] -= np.mean(x_img[:,:,1])
x_img[:,:,2] -= np.mean(x_img[:,:,2])
# C.img_scaling_factor = 1
x_img /= C.img_scaleing_factor
## 1， channel, height , weight
x_img = np.transpose(x_img, (2,0,1))
x_img = np.expand_dims(x_img, axis = 0)
C.std_scaling = 4
## 1, num_anchor*4*2, height, weight
## y_rpn_reg.shape[1] // 2 = 9
## y_rpn_overlap 前面9个
## y_rpn_reg 后面9个
y_rpn_reg[:, y_rpn_reg.shape[1] // 2:, :, :] *= C.std_scaling
## 1，height width channel for tensorflow form
x_img = np.transpose(x_img, (0,2,3,1))
## 1，height, width, num_anchor*x for tensorflow form
y_rpn_cls = np.transpose(y_rpn_cls, (0,2,3,1))
y_rpn_reg = np.transpose(y_rpn_reg, (0,2,3,1))
#build a generator
yield np.copy(x_img), [np.copy(y_rpn_cls), np.copy(y_rpn_reg)], img_data_aug
#keep moving
except Exception as e:
print(e)
continue