def gen(self):
while True:
# 一个epoch重新打乱
np.random.shuffle(self.annotation_list)
for n in range(self.size // self.batch_size):
images = np.zeros((self.batch_size,) + self.input_shape, dtype=np.float32)
image_metas = np.zeros((self.batch_size, 12), dtype=np.float32)
batch_gt_boxes = np.zeros((self.batch_size, self.max_gt_num, 5), dtype=np.float32)
batch_gt_class_ids = np.ones((self.batch_size, self.max_gt_num, 2), dtype=np.uint8)
# 随机选择
indices = np.random.choice(self.size, self.batch_size, replace=False)
for i, index in enumerate(indices):
# 加载图像
image = image_utils.load_image(self.annotation_list[index]['filepath'])
# 数据增广:水平翻转、随机裁剪
gt_boxes = self.annotation_list[index]['boxes'].copy() # 不改变原来的
if self.horizontal_flip and random.random() > 0.5:
image, gt_boxes = image_flip(image, gt_boxes)
if self.random_crop and random.random() > 0.5:
image, gt_boxes = image_crop(image, gt_boxes)
# resize图像
images[i], image_metas[i], gt_boxes = image_utils.resize_image_and_gt(image,
self.input_shape[0],
gt_boxes)
# pad gt到固定个数
batch_gt_boxes[i] = np_utils.pad_to_fixed_size(gt_boxes, self.max_gt_num)
batch_gt_class_ids[i] = np_utils.pad_to_fixed_size(
np.expand_dims(self.annotation_list[index]['labels'], axis=1),
self.max_gt_num)
yield {"input_image": images,
"input_image_meta": image_metas,
"input_gt_boxes": batch_gt_boxes,
"input_gt_class_ids": batch_gt_class_ids}, None
def _show_inference(id, ax=None):
image = image_utils.load_image(all_img_info[id]['filepath'])
image, image_meta, _ = image_utils.resize_image_and_gt(image,
config.IMAGE_MAX_DIM,
all_img_info[id]['boxes'])
boxes, scores, class_ids, _, image_metas = m.predict(
[np.expand_dims(image, axis=0), np.expand_dims(image_meta, axis=0)])
boxes = np_utils.remove_pad(boxes[0])
scores = np_utils.remove_pad(scores[0])[:, 0]
class_ids = np_utils.remove_pad(class_ids[0])[:, 0]
visualize.display_instances(image, boxes[:5],
class_ids[:5],
id_mapping,
scores=scores[:5],
ax=ax)
def gen_val(self):
"""
评估生成器
:return:
"""
for idx, image_info in enumerate(self.annotation_list):
# 加载图像
image = image_utils.load_image(self.annotation_list[idx]['filepath'])
image, image_meta, _ = image_utils.resize_image_and_gt(image,
self.input_shape[0])
if idx % 200 == 0:
print("开始预测:{}张图像".format(idx))
yield {"input_image": np.asarray([image]),
"input_image_meta": np.asarray([image_meta])}
def __getitem__(self, index):
start = index * self.batch_size
end = min((index + 1) * self.batch_size, self.size)
indices = np.arange(start, end)
images = np.zeros((end - start, ) + self.input_shape, dtype=np.float32)
image_metas = np.zeros((end - start, 12), dtype=np.float32)
for i, index in enumerate(indices):
# 加载图像
image = image_utils.load_image(
self.annotation_list[index]['filepath'])
images[i], image_metas[i], _ = image_utils.resize_image_and_gt(
image, self.input_shape[0])
if index % 200 == 0:
print("开始预测:{}张图像".format(index))
images = np.asarray(images, np.float32) - self.mean_pixel # 减去均值
return {"input_image": images, "input_image_meta": image_metas}
def __getitem__(self, index):
indices = np.arange(index * self.batch_size,
(index + 1) * self.batch_size)
images = np.zeros((self.batch_size, ) + self.input_shape,
dtype=np.float32)
image_metas = np.zeros((self.batch_size, 12), dtype=np.float32)
batch_gt_boxes = np.zeros((self.batch_size, self.max_gt_num, 5),
dtype=np.float32)
batch_gt_class_ids = np.ones((self.batch_size, self.max_gt_num, 2),
dtype=np.uint8)
for i, index in enumerate(indices):
# 加载图像
image = image_utils.load_image(
self.annotation_list[index]['filepath'])
# 数据增广:水平翻转、随机裁剪
gt_boxes = self.annotation_list[index]['boxes'].copy() # 不改变原来的
if self.horizontal_flip and random.random() > 0.5:
image, gt_boxes = image_flip(image, gt_boxes)
if self.random_crop and random.random() > 0.5:
image, gt_boxes = image_crop(image, gt_boxes)
# resize图像
images[i], image_metas[
i], gt_boxes = image_utils.resize_image_and_gt(
image, self.input_shape[0], gt_boxes)
# pad gt到固定个数
batch_gt_boxes[i] = np_utils.pad_to_fixed_size(
gt_boxes, self.max_gt_num)
batch_gt_class_ids[i] = np_utils.pad_to_fixed_size(
np.expand_dims(self.annotation_list[index]['labels'], axis=1),
self.max_gt_num)
images = np.asarray(images, np.float32) - self.mean_pixel # 减去均值
return {
"input_image": images,
"input_image_meta": image_metas,
"input_gt_boxes": batch_gt_boxes,
"input_gt_class_ids": batch_gt_class_ids
}, None