def __getitem__(self, idx):
'''Load the image and its bboxes for the given index.
Args
---
idx: the index of images.
Returns
---
tuple: A tuple containing the following items: image,
bboxes, labels.
'''
img_info = self.img_infos[idx]
ann_info = self._load_ann_info(idx)
# load the image.
img = cv2.imread(f"{self.image_dir}/{img_info['file_name']}",
cv2.IMREAD_COLOR)
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
ori_shape = img.shape
# Load the annotation.
ann = self._parse_ann_info(ann_info)
bboxes = ann['bboxes']
labels = ann['labels']
segments = ann['segments'] # list
masks, global_mask = self._segments_to_mask(segments, labels,
ori_shape[:2])
flip = True if np.random.rand() < self.flip_ratio else False
# Handle the image
img, img_shape, scale_factor, masks, global_mask = self.img_transform(
img, masks, global_mask, flip)
# Resize masks to smaller size to reduce memory usage
if self.config.USE_MINI_MASK and not self.debug:
masks = self.minimize_mask(bboxes,
masks,
self.config.MINI_MASK_SHAPE,
softmask=self.config.SOFT_MASK)
pad_shape = img.shape
# Handle the annotation.
bboxes, labels = self.bbox_transform(bboxes, labels, img_shape,
scale_factor, flip)
# Handle the meta info.
img_meta_dict = dict({
'ori_shape': ori_shape,
'img_shape': img_shape,
'pad_shape': pad_shape,
'scale_factor': scale_factor,
'flip': flip
})
img_meta = utils.compose_image_meta(img_meta_dict)
return img, img_meta, bboxes, labels, masks, global_mask
def __getitem__(self, idx):
'''Load the image and its bboxes for the given index.
Args
---
idx: the index of images.
Returns
---
tuple: A tuple containing the following items: image,
bboxes, labels.
'''
img_info = self.img_infos[idx]
ann_info = self._load_ann_info(idx)
# load the image.
img = cv2.imread(osp.join(self.image_dir, img_info['file_name']),
cv2.IMREAD_COLOR)
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
ori_shape = img.shape
# Load the annotation.
ann = self._parse_ann_info(ann_info)
bboxes = ann['bboxes']
labels = ann['labels']
flip = True if np.random.rand() < self.flip_ratio else False
# Handle the image
img, img_shape, scale_factor = self.img_transform(img, flip)
pad_shape = img.shape
# Handle the annotation.
bboxes, labels = self.bbox_transform(bboxes, labels, img_shape,
scale_factor, flip)
# Handle the meta info.
img_meta_dict = dict({
'ori_shape': ori_shape,
'img_shape': img_shape,
'pad_shape': pad_shape,
'scale_factor': scale_factor,
'flip': flip
})
img_meta = utils.compose_image_meta(img_meta_dict)
return img, img_meta, bboxes, labels
def __getitem__(self, idx):
"""根据下标号返回图片和标注信息
Args
---
idx: 图片下标
Returns
---
img: [height, width, channels]
img_meta: [11]
bboxes: [num_boxes, 4]
labels: [num_boxes]
"""
# img_infos: dict dict为coco格式的图片信息
# ann_info: [dict] dict为coco格式的标注信息
img_info = self.img_infos[idx]
ann_info = self._load_ann_info(idx)
# cv2加载BGR图片，转换成RGB
img = cv2.imread(osp.join(self.image_dir, img_info['file_name']), cv2.IMREAD_COLOR)
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
ori_shape = img.shape
# ann: dict 字典的key为bboxes, labels, bboxes_ignore
ann = self._parse_ann_info(ann_info)
bboxes = ann['bboxes']
labels = ann['labels']
flip = True if np.random.rand() < self.flip_ratio else False # 按概率随机决定是否翻转
# 图片和边界框的预处理
img, img_shape, scale_factor = self.img_transform(img, flip)
bboxes = self.bbox_transform(bboxes, img_shape, scale_factor, flip)
# 整合图片信息
img_meta_dict = dict({
'ori_shape': ori_shape, # 图片原始尺寸
'img_shape': img_shape, # 图片缩放后的尺寸
'pad_shape': img.shape, # 图片缩放+填充后的尺寸
'scale_factor': scale_factor, # 缩放因子，缩放后尺寸 = int(缩放前尺寸*scale_factor + 0.5)
'flip': flip # 是否进行左右翻转
})
img_meta = utils.compose_image_meta(img_meta_dict)
return img, img_meta, bboxes, labels
source_class_ids = self.source_class_ids[self.image_info[image_id]["source"]]
active_class_ids[source_class_ids] = 1
# Resize masks to smaller size to reduce memory usage
if use_mini_mask:
mask = utils.minimize_mask(bbox, mask, config.MINI_MASK_SHAPE)
# Image meta data
img_meta_dict = dict({
'ori_shape': original_shape,
'img_shape': ,
'pad_shape': image.shape,
'scale_factor': scale,
'flip': flip
})
img_meta = utils.compose_image_meta(img_meta_dict)
image_meta = compose_image_meta(image_id, original_shape, image.shape,
window, scale, active_class_ids)
return image, image_meta, class_ids, bbox, mask
# def compose_image_meta(image_id, original_image_shape, image_shape,
# window, scale, active_class_ids):
# """Takes attributes of an image and puts them in one 1D array.
# image_id: An int ID of the image. Useful for debugging.
# original_image_shape: [H, W, C] before resizing or padding.
# image_shape: [H, W, C] after resizing and padding
# window: (y1, x1, y2, x2) in pixels. The area of the image where the real
# image is (excluding the padding)
# scale: The scaling factor applied to the original image (float32)