def preprocess(img, bbox_labels, mode, settings):
img_width, img_height = img.size
sampled_labels = bbox_labels
if mode == 'train':
if settings._apply_distort:
img = image_util.distort_image(img, settings)
if settings._apply_expand:
img, bbox_labels, img_width, img_height = image_util.expand_image(
img, bbox_labels, img_width, img_height, settings)
# sampling
batch_sampler = []
# hard-code here
batch_sampler.append(
image_util.sampler(1, 1, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0))
batch_sampler.append(
image_util.sampler(1, 50, 0.3, 1.0, 0.5, 2.0, 0.1, 0.0))
batch_sampler.append(
image_util.sampler(1, 50, 0.3, 1.0, 0.5, 2.0, 0.3, 0.0))
batch_sampler.append(
image_util.sampler(1, 50, 0.3, 1.0, 0.5, 2.0, 0.5, 0.0))
batch_sampler.append(
image_util.sampler(1, 50, 0.3, 1.0, 0.5, 2.0, 0.7, 0.0))
batch_sampler.append(
image_util.sampler(1, 50, 0.3, 1.0, 0.5, 2.0, 0.9, 0.0))
batch_sampler.append(
image_util.sampler(1, 50, 0.3, 1.0, 0.5, 2.0, 0.0, 1.0))
sampled_bbox = image_util.generate_batch_samples(
batch_sampler, bbox_labels)
img = np.array(img)
if len(sampled_bbox) > 0:
idx = int(np.random.uniform(0, len(sampled_bbox)))
img, sampled_labels = image_util.crop_image(
img, bbox_labels, sampled_bbox[idx], img_width, img_height)
img = Image.fromarray(img)
img = img.resize((settings.resize_w, settings.resize_h), Image.ANTIALIAS)
img = np.array(img)
if mode == 'train':
mirror = int(np.random.uniform(0, 2))
if mirror == 1:
img = img[:, ::-1, :]
for i in six.moves.xrange(len(sampled_labels)):
tmp = sampled_labels[i][1]
sampled_labels[i][1] = 1 - sampled_labels[i][3]
sampled_labels[i][3] = 1 - tmp
# HWC to CHW
if len(img.shape) == 3:
img = np.swapaxes(img, 1, 2)
img = np.swapaxes(img, 1, 0)
# RBG to BGR
img = img[[2, 1, 0], :, :]
img = img.astype('float32')
img -= settings.img_mean
img = img * 0.007843
return img, sampled_labels
def reader():
with open(file_list) as flist:
lines = [line.strip() for line in flist]
if shuffle:
# 打乱数据
random.shuffle(lines)
for line in lines:
if mode == 'train' or mode == 'test':
# 获取图像的路径和对应标注的文件路径
img_path, label_path = line.split()
img_path = os.path.join(settings.data_dir, img_path)
label_path = os.path.join(settings.data_dir, label_path)
elif mode == 'infer':
# 获取图像的路径
img_path = os.path.join(settings.data_dir, line)
print img_path
img = Image.open(img_path)
# 获取图像的原始大小
img_width, img_height = img.size
img = np.array(img)
# 保存列表的结构: label | xmin | ymin | xmax | ymax | difficult
if mode == 'train' or mode == 'test':
# 保存每个标注框
bbox_labels = []
# 开始读取标注信息
root = xml.etree.ElementTree.parse(label_path).getroot()
# 查询每个标注的信息
for object in root.findall('object'):
# 每个标注框的信息
bbox_sample = []
# start from 1
bbox_sample.append(
float(
settings.label_list.index(
object.find('name').text)))
bbox = object.find('bndbox')
difficult = float(object.find('difficult').text)
# 获取标注信息，并计算比例保存
bbox_sample.append(
float(bbox.find('xmin').text) / img_width)
bbox_sample.append(
float(bbox.find('ymin').text) / img_height)
bbox_sample.append(
float(bbox.find('xmax').text) / img_width)
bbox_sample.append(
float(bbox.find('ymax').text) / img_height)
bbox_sample.append(difficult)
# 将整个框的信息保存
bbox_labels.append(bbox_sample)
sample_labels = bbox_labels
if mode == 'train':
batch_sampler = []
# 样本定义
batch_sampler.append(
image_util.sampler(1, 1, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0))
batch_sampler.append(
image_util.sampler(1, 50, 0.3, 1.0, 0.5, 2.0, 0.1, 0.0))
batch_sampler.append(
image_util.sampler(1, 50, 0.3, 1.0, 0.5, 2.0, 0.3, 0.0))
batch_sampler.append(
image_util.sampler(1, 50, 0.3, 1.0, 0.5, 2.0, 0.5, 0.0))
batch_sampler.append(
image_util.sampler(1, 50, 0.3, 1.0, 0.5, 2.0, 0.7, 0.0))
batch_sampler.append(
image_util.sampler(1, 50, 0.3, 1.0, 0.5, 2.0, 0.9, 0.0))
batch_sampler.append(
image_util.sampler(1, 50, 0.3, 1.0, 0.5, 2.0, 0.0, 1.0))
# 将标注信息生成一个batch
sampled_bbox = image_util.generate_batch_samples(batch_sampler,
bbox_labels)
if len(sampled_bbox) > 0:
idx = int(random.uniform(0, len(sampled_bbox)))
img, sample_labels = image_util.crop_image(
img, bbox_labels, sampled_bbox[idx], img_width,
img_height)
img = Image.fromarray(img)
# 设置图像大小
img = img.resize((settings.resize_w, settings.resize_h),
Image.ANTIALIAS)
img = np.array(img)
if mode == 'train':
mirror = int(random.uniform(0, 2))
if mirror == 1:
img = img[:, ::-1, :]
for i in xrange(len(sample_labels)):
tmp = sample_labels[i][1]
sample_labels[i][1] = 1 - sample_labels[i][3]
sample_labels[i][3] = 1 - tmp
if len(img.shape) == 3:
img = np.swapaxes(img, 1, 2)
img = np.swapaxes(img, 1, 0)
img = img.astype('float32')
img -= settings.img_mean
img = img.flatten()
if mode == 'train' or mode == 'test':
if mode == 'train' and len(sample_labels) == 0:
continue
yield img.astype('float32'), sample_labels
elif mode == 'infer':
yield img.astype('float32')
def preprocess(img, bbox_labels, mode, settings, image_path):
img_width, img_height = img.size
sampled_labels = bbox_labels
if mode == 'train':
if settings.apply_distort:
img = image_util.distort_image(img, settings)
if settings.apply_expand:
img, bbox_labels, img_width, img_height = image_util.expand_image(
img, bbox_labels, img_width, img_height, settings)
# sampling
batch_sampler = []
# used for continuous evaluation
if 'ce_mode' in os.environ:
random.seed(0)
np.random.seed(0)
prob = np.random.uniform(0., 1.)
if prob > settings.data_anchor_sampling_prob:
scale_array = np.array([16, 32, 64, 128, 256, 512])
batch_sampler.append(
image_util.sampler(1, 10, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.2,
0.0, True))
sampled_bbox = image_util.generate_batch_random_samples(
batch_sampler, bbox_labels, img_width, img_height, scale_array,
settings.resize_width, settings.resize_height)
img = np.array(img)
if len(sampled_bbox) > 0:
idx = int(np.random.uniform(0, len(sampled_bbox)))
img, sampled_labels = image_util.crop_image_sampling(
img, bbox_labels, sampled_bbox[idx], img_width, img_height,
settings.resize_width, settings.resize_height,
settings.min_face_size)
img = img.astype('uint8')
img = Image.fromarray(img)
else:
# hard-code here
batch_sampler.append(
image_util.sampler(1, 50, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0,
0.0, True))
batch_sampler.append(
image_util.sampler(1, 50, 0.3, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0,
0.0, True))
batch_sampler.append(
image_util.sampler(1, 50, 0.3, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0,
0.0, True))
batch_sampler.append(
image_util.sampler(1, 50, 0.3, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0,
0.0, True))
batch_sampler.append(
image_util.sampler(1, 50, 0.3, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0,
0.0, True))
sampled_bbox = image_util.generate_batch_samples(
batch_sampler, bbox_labels, img_width, img_height)
img = np.array(img)
if len(sampled_bbox) > 0:
idx = int(np.random.uniform(0, len(sampled_bbox)))
img, sampled_labels = image_util.crop_image(
img, bbox_labels, sampled_bbox[idx], img_width, img_height,
settings.resize_width, settings.resize_height,
settings.min_face_size)
img = Image.fromarray(img)
interp_mode = [
Image.BILINEAR, Image.HAMMING, Image.NEAREST, Image.BICUBIC,
Image.LANCZOS
]
interp_indx = np.random.randint(0, 5)
img = img.resize((settings.resize_width, settings.resize_height),
resample=interp_mode[interp_indx])
img = np.array(img)
if mode == 'train':
mirror = int(np.random.uniform(0, 2))
if mirror == 1:
img = img[:, ::-1, :]
for i in six.moves.xrange(len(sampled_labels)):
tmp = sampled_labels[i][1]
sampled_labels[i][1] = 1 - sampled_labels[i][3]
sampled_labels[i][3] = 1 - tmp
img = to_chw_bgr(img)
img = img.astype('float32')
img -= settings.img_mean
img = img * settings.scale
return img, sampled_labels
def reader():
lines = open(file_list, "r").readlines()
if shuffle:
random.shuffle(lines)
for line_id, line in enumerate(lines):
if mode == "train" or mode == "test":
img_path, label_path = line.split()
img_path = os.path.join(settings.data_dir, "resized_images",
img_path)
label_path = os.path.join(settings.data_dir, "annotations",
label_path)
elif mode == "infer":
img_path = os.path.join(settings.data_dir, "resized_images",
line.strip().split()[0])
img = Image.open(img_path)
img_width, img_height = img.size
img = np.array(img)
if mode == "train" or mode == "test":
labels = json.loads(open(label_path, "r").readline())
if len(labels) == 0: continue
bbox_labels = []
for lbl in labels:
bbox_labels.append([
settings.label_list[lbl["text"]], # label
lbl["posX"] / ORIGINAL_WIDTH, # xmin
lbl["posY"] / ORIGINAL_HEIGH, # ymin
(lbl["posX"] + lbl["width"]) / ORIGINAL_WIDTH, # xmax
(lbl["posY"] + lbl["height"]) / ORIGINAL_HEIGH, # ymax
1. # difficult
])
sample_labels = bbox_labels
if mode == "train":
sampled_bbox = image_util.generate_batch_samples(
get_batch_sampler(), bbox_labels, img_width,
img_height)
if len(sampled_bbox) > 0:
idx = int(random.uniform(0, len(sampled_bbox)))
img, sample_labels = image_util.crop_image(
img, bbox_labels, sampled_bbox[idx], img_width,
img_height)
img = Image.fromarray(img)
img = img.resize((settings.resize_w, settings.resize_h),
Image.ANTIALIAS)
img = np.array(img)
if mode == "train":
mirror = int(random.uniform(0, 2))
if mirror == 1:
img = img[:, ::-1, :]
for i in xrange(len(sample_labels)):
tmp = sample_labels[i][1]
sample_labels[i][1] = 1 - sample_labels[i][3]
sample_labels[i][3] = 1 - tmp
img = np.transpose(img, [2, 0, 1]) # channel first
img = img.astype("float32")
img -= settings.img_mean
img = img.flatten()
if mode == "train" or mode == "test":
if mode == "train" and len(sample_labels) == 0: continue
yield img, sample_labels
elif mode == "infer":
yield img.astype("float32")