def _get_image_blob(roidb, scale_inds):
"""Builds an input blob from the images in the roidb at the specified
scales.
"""
num_images = len(roidb)
processed_ims = []
im_scales = []
for i in range(num_images):
#im = cv2.imread(roidb[i]['image'])
im = imread(roidb[i]['image'])
if len(im.shape) == 2:
im = im[:, :, np.newaxis]
im = np.concatenate((im, im, im), axis=2)
# flip the channel, since the original one using cv2
# rgb -> bgr
im = im[:, :, ::-1]
if roidb[i]['flipped']:
im = im[:, ::-1, :]
target_size = cfg.TRAIN.SCALES[scale_inds[i]]
im, im_scale = prep_im_for_blob(im, cfg.PIXEL_MEANS, target_size,
cfg.TRAIN.MAX_SIZE)
im_scales.append(im_scale)
processed_ims.append(im)
# Create a blob to hold the input images
blob = im_list_to_blob(processed_ims)
return blob, im_scales
def _get_image_blob(im):
"""Converts an image into a network input.
Arguments:
im (ndarray): a color image in BGR order
Returns:
blob (ndarray): a data blob holding an image pyramid
im_scale_factors (list): list of image scales (relative to im) used
in the image pyramid
"""
im_orig = im.astype(np.float32, copy=True)
im_orig -= cfg.PIXEL_MEANS
im_shape = im_orig.shape
processed_ims = []
im_scale_factors = []
size = cfg.TEST.SIZE
im_scale_w = float(size) / float(im_shape[1])
im_scale_h = float(size) / float(im_shape[0])
# Prevent the biggest axis from being more than MAX_SIZE
im = cv2.resize(im_orig,
(size,size),
interpolation=cv2.INTER_LINEAR)
im_scale_factors.append(im_scale_w)
im_scale_factors.append(im_scale_h)
processed_ims.append(im)
# Create a blob to hold the input images
blob = im_list_to_blob(processed_ims)
return blob, np.array(im_scale_factors)
def _get_image_blob(im):
"""Converts an image into a network input.
Arguments:
im (ndarray): a color image in BGR order
Returns:
blob (ndarray): a data blob holding an image pyramid
im_scale_factors (list): list of image scales (relative to im) used
in the image pyramid
"""
im_orig = im.astype(np.float32, copy=True)
im_orig -= cfg.PIXEL_MEANS
im_shape = im_orig.shape
im_size_min = np.min(im_shape[0:2])
im_size_max = np.max(im_shape[0:2])
processed_ims = []
im_scale_factors = []
for target_size in cfg.TEST.SCALES:
im_scale = float(target_size) / float(im_size_min)
# Prevent the biggest axis from being more than MAX_SIZE
if np.round(im_scale * im_size_max) > cfg.TEST.MAX_SIZE:
im_scale = float(cfg.TEST.MAX_SIZE) / float(im_size_max)
im = cv2.resize(im_orig, None, None, fx=im_scale, fy=im_scale,
interpolation=cv2.INTER_LINEAR)
im_scale_factors.append(im_scale)
processed_ims.append(im)
# Create a blob to hold the input images
blob = im_list_to_blob(processed_ims)
return blob, np.array(im_scale_factors)
def _get_image_blob(roidb, scale_inds):
"""Builds an input blob from the images in the roidb at the specified
scales.
"""
num_images = len(roidb) # num_images = 1
processed_ims = []
im_scales = []
for i in range(num_images):
im = imread(roidb[i]['image'])
if len(im.shape) == 2:
im = im[:, :, np.newaxis]
im = np.concatenate((im, im, im), axis=2)
# flip the channel, since the original one using cv2
# rgb -> bgr
im = im[:, :, ::-1]
if roidb[i]['flipped']:
im = im[:, ::-1, :] # 对图像进行水平翻转
target_size = cfg.TRAIN.SCALES[scale_inds[i]]
im, im_scale = prep_im_for_blob(im, cfg.PIXEL_MEANS, target_size,
cfg.TRAIN.MAX_SIZE)
# im_scale = (target_size) / float(im_size_min)，表示原始图像的短边到训练尺寸600的变换倍数
im_scales.append(im_scale)
processed_ims.append(im)
# Create a blob to hold the input images
blob = im_list_to_blob(processed_ims)
# 返回blob形式[1,w,h,c]，im_scales表示图像resize的倍数
return blob, im_scales
def get_evaluate_batch(self, im_path, index):
# Sample random scales to use for each image in this batch
# Get the input image blob, formatted for caffe
# im_blob, im_scales = _get_image_blob(roidb, random_scale_inds)
im = imread(im_path)
if len(im.shape) == 2:
im = im[:, :, np.newaxis]
im = np.concatenate((im, im, im), axis=2)
# flip the channel, since the original one using cv2
# rgb -> bgr
im = im[:, :, ::-1]
target_size = cfg.TRAIN.SCALES[0]
im, im_scale = prep_im_for_blob(im, cfg.PIXEL_MEANS, target_size,
cfg.TRAIN.MAX_SIZE)
im_blob = im_list_to_blob([im])
blobs = {'data': im_blob}
# gt boxes: (x1, y1, x2, y2, cls)
gt_boxes = np.empty((0, 5), dtype=np.float32)
blobs['gt_boxes'] = gt_boxes
blobs['im_info'] = np.array([[im.shape[0], im.shape[1], im_scale]], dtype=np.float32)
blobs['img_id'] = index
return blobs
def _get_image_blob(im):
"""将一幅图像转化为网络需要的输入
Arguments:输入一个通道顺序为BGR的图像
im (ndarray): a color image in BGR order
Returns:返回一个图像金字塔列表
blob (ndarray): a data blob holding an image pyramid
im_scale_factors (list): list of image scales (relative to im) used
in the image pyramid
"""
im_orig = im.astype(np.float32, copy=True)
#减去数据训练集的统计平均值，来消除公共的部分，以凸显个体之间的特征和差异
im_orig -= cfg.PIXEL_MEANS
im_shape = im_orig.shape
im_size_min = np.min(im_shape[0:2])
im_size_max = np.max(im_shape[0:2])
processed_ims = []
im_scale_factors = []
for target_size in cfg.TEST.SCALES:
# 限制最小边为600，最大边为1000，对于输入图像优先考虑最大边的限制
# 输入图像的大小是：375*500*3，则resize后的图像大小为：600*800*3
# 输入图像的大小是：375*800*3，则resize后的图像大小为：469*1000*3
im_scale = float(target_size) / float(im_size_min)
# Prevent the biggest axis from being more than MAX_SIZE
if np.round(
im_scale * im_size_max) > cfg.TEST.MAX_SIZE: #np.round返回四舍五入值
im_scale = float(cfg.TEST.MAX_SIZE) / float(im_size_max)
#resize参数：输入图像、输出图像、输出尺寸、w方向缩放因子、h方向…、插值方法
im = cv2.resize(im_orig,
None,
None,
fx=im_scale,
fy=im_scale,
interpolation=cv2.INTER_LINEAR)
im_scale_factors.append(im_scale)
processed_ims.append(im)
#Create a blob to hold the input images
blob = im_list_to_blob(processed_ims)
return blob, np.array(im_scale_factors)
