def parse_args():
"""
处理脚本参数，支持相对路径
img_file 文件路径，默认文件夹：img_downloader/urls
out_folder 输出文件夹，默认文件夹：img_data
:return: arg_img，文件路径；out_folder，输出文件夹
"""
parser = argparse.ArgumentParser(description='下载数据脚本')
parser.add_argument('--img_file', required=True, help='文件路径', type=str)
parser.add_argument('--out_folder', help='输出文件夹', type=str)
args = parser.parse_args()
arg_img = args.img_file
if len(arg_img.split('/')) == 1:
arg_img = os.path.join(ROOT_DIR, 'img_downloader', 'urls', arg_img)
print_info("文件路径：%s" % arg_img)
arg_out = args.out_folder
if not arg_out:
file_name = arg_img.split('/')[-1]
arg_out = os.path.join(ROOT_DIR, 'img_data', file_name)
elif len(arg_out.split('/')) == 1:
arg_out = os.path.join(ROOT_DIR, 'img_data', arg_out)
print_info("输出文件夹：%s" % arg_out)
return arg_img, arg_out
def format_img_and_anno(img_folder):
"""
格式化输出。图片和标注文件夹
:param img_folder: 图片文件夹
:return:
"""
file_paths, file_names = traverse_dir_files(img_folder)
img_dict = dict() # 将标注和图片路径，生成一个字典
for file_path, file_name in zip(file_paths, file_names):
if file_name.endswith('.jpg'):
name = file_name.replace('.jpg', '')
if name not in img_dict:
img_dict[name] = (None, None)
(img_p, anno_p) = img_dict[name]
img_dict[name] = (file_path, anno_p)
if file_name.endswith('.xml'):
name = file_name.replace('.xml', '')
if name not in img_dict:
img_dict[name] = (None, None)
(img_p, anno_p) = img_dict[name]
img_dict[name] = (img_p, file_path)
print_info('图片数: {}'.format(len(img_dict.keys())))
return img_dict
def detect_img(self, yolo, img_path, anno_path, out_folder=None):
"""
检测图片，与标注对比
:param yolo: YOLO
:param img_path: 图片路径
:param anno_path: 标注路径
:param out_folder: 存储错误图像
:return:
"""
img_data = Image.open(img_path)
boxes, scores, classes = yolo.__detect_img_facets(img_data)
boxes = self.reform_boxes(boxes)
print_info('检测: {}'.format(boxes))
anno_boxes, name_list = read_anno_xml(anno_path)
anno_boxes = self.reform_boxes2(anno_boxes)
print_info('真值: {}'.format(anno_boxes))
precision, recall = self.iou_of_boxes(boxes, anno_boxes)
if True or out_folder and (precision != 1.0 or recall != 1.0):
img_data = yolo.draw_boxes(img_data, boxes, scores, classes,
[(255, 0, 0)])
img_data = yolo.draw_boxes(img_data, anno_boxes,
[1.0 for i in range(len(anno_boxes))],
[0 for i in range(len(anno_boxes))],
[(128, 128, 255)])
# 图片的缩略图
# size = 1024, 1024
# img_data.thumbnail(size, Image.ANTIALIAS)
img_name = img_path.split('/')[-1]
out_img = os.path.join(out_folder, img_name + '.d.png')
img_data.save(out_img) # 存储
# img_data.show() # 显示
return img_path, precision, recall
def detect_img_folder(img_folder, out_folder, yolo):
mkdir_if_not_exist(out_folder)
path_list, name_list = traverse_dir_files(img_folder)
print_info('图片数: %s' % len(path_list))
_, imgs_names = traverse_dir_files(out_folder)
count = 0
for path, name in zip(path_list, name_list):
if path.endswith('.gif'):
continue
out_name = name + '.d.jpg'
if out_name in imgs_names:
print_info('已检测: %s' % name)
continue
print_info('检测图片: %s' % name)
try:
image = Image.open(path)
out_file = os.path.join(ROOT_DIR, 'face', 'yolov3', 'output_data',
'logAll_res.txt')
r_image = yolo.detect_image(image, ('logAll/' + name), out_file)
r_image.save(os.path.join(out_folder, name + '.d.jpg'))
except Exception as e:
print(e)
pass
count += 1
if count % 100 == 0:
print_info('已检测: %s' % count)
yolo.close_session()
def __init__(self,
model_path='model_data/ep074-loss26.535-val_loss27.370.h5',
classes_path='configs/wider_classes.txt',
anchors_path='configs/yolo_anchors.txt'):
self.model_path = model_path # 模型文件
self.classes_path = classes_path # 类别文件
self.anchors_path = anchors_path # anchors文件
print_info('模型: {}'.format(model_path))
self.score = 0.25
self.iou = 0.30
print_info('置信度: {}, IoU: {}'.format(self.score, self.iou)) # 输出参数
self.class_names = self._get_class() # 获取类别
self.anchors = self._get_anchors() # 获取anchor
self.sess = K.get_session()
print_info('anchors: {}, 类别: {}'.format(len(self.anchors),
len(self.class_names)))
self.model_image_size = (416, 416) # fixed size or (None, None), hw
print_info('图片尺寸: {}'.format(self.model_image_size)) # 输出参数
self.yolo_model = None
self.colors = None
self.input_image_shape = None
self.boxes, self.scores, self.classes = self.generate()
def iou_of_boxes(boxes1, boxes2):
"""
框的IOU
:param boxes1_: 真值
:param boxes2_: 预测
:return: 精准和召回
"""
res_list = []
boxes1_ = copy.deepcopy(boxes1)
boxes2_ = copy.deepcopy(boxes2)
t1, t2 = len(boxes1_), len(boxes2_)
for box1 in boxes1_:
final_iou = 0
final_box = None
for box2 in boxes2_:
iou = bb_intersection_over_union(box1, box2)
if iou > final_iou:
final_iou = iou
final_box = box2
if final_iou > 0.5:
res_list.append((box1, final_box, final_iou))
boxes2_.remove(final_box)
tr = len(res_list)
if tr == 0: # 没有检测源, 则认为正确
if t1 == 0 and t2 != 0:
recall, precision = 1.0, 0.0
elif t1 != 0 and t2 == 0:
recall, precision = 0.0, 1.0
elif t1 != 0 and t2 != 0:
recall, precision = 0.0, 0.0
else:
recall, precision = 1.0, 1.0
else:
recall = safe_div(tr, t1) # 召回率
precision = safe_div(tr, t2) # 精准率
print_info('精准: {:.4f}, 召回: {:.4f}, 匹配结果: {}'.format(
precision, recall, res_list))
return [precision, recall]
def load_model(self):
net = DarkNet(input_dim=self.input_dim,
num_classes=self.num_classes) # 基础网络DarkNet
net.initialize(ctx=self.ctx)
if self.params_path.endswith(".params"): # 加载模型
net.load_params(self.params_path)
elif self.params_path.endswith(".weights"):
tmp_batch = nd.uniform(shape=(1, 3, self.input_dim,
self.input_dim),
ctx=self.ctx)
net(tmp_batch)
net.load_weights(self.params_path, fine_tune=False)
else:
print("params {} load error!".format(self.params_path))
exit()
print_info("加载参数: {}".format(self.params_path))
net.hybridize()
return net
def download_imgs_for_mp(img_file, out_folder, prefix=None, n_prc=10):
"""
多线程下载
:param img_file: 图片文件
:param out_folder: 输出文件夹
:param prefix: 图片前缀
:param n_prc: 进程数, 默认40个
:return: None
"""
print_info('进程总数: %s' % n_prc)
pool = Pool(processes=n_prc) # 多线程下载
paths_list = read_file(img_file)
print_info('文件数: %s' % len(paths_list))
_, imgs_names = traverse_dir_files(out_folder)
for (index, path) in enumerate(paths_list):
if prefix:
pool.apply_async(download_img,
(path, out_folder, imgs_names,
prefix + '_' + str(index) + '.jpg'))
else:
pool.apply_async(download_img, (path, out_folder, imgs_names))
pool.close()
pool.join()
# _, imgs_names = traverse_dir_files(out_folder)
# print_info('图片总数: %s' % len(imgs_names))
print_info('全部下载完成')
def verify_model(self):
"""
处理标记文件夹
:param img_folder: 图片文件夹
:param out_folder:
:return:
"""
yolo = YoloV3(model_path=self.model_path,
classes_path=self.classes_path,
anchors_path=self.anchors_path) # yolo算法类
img_dict = format_img_and_anno(self.img_folder)
res_dict = dict()
for count, img_name in enumerate(img_dict):
print_info('-' * 50)
print_info('图片: {}'.format(img_name))
(img_p, anno_p) = img_dict[img_name]
_, precision, recall = self.detect_img(yolo, img_p, anno_p,
self.out_folder)
res_dict[img_name] = (precision, recall)
if count == 10:
break
ap, ar = 0, 0
for name in res_dict.keys():
precision, recall = res_dict[name]
ap += precision
ar += recall
mAp = safe_div(ap, len(res_dict.keys()))
mAr = safe_div(ar, len(res_dict.keys()))
print_info('平均精准率: {:.4f} %, 平均召回率: {:.4f} %'.format(
mAp * 100, mAr * 100))
def __load_model(self):
"""
加载网络模型
:return: 网络
"""
net = DarkNet(input_dim=self.input_dim, num_classes=self.num_classes) # 基础网络 DarkNet
net.initialize(ctx=self.ctx) # 网络环境 cpu or gpu
print_info("模型: {}".format(self.params_path))
if self.params_path.endswith(".params"): # 加载参数
net.load_params(self.params_path)
elif self.params_path.endswith(".weights"): # 加载模型
tmp_batch = nd.uniform(shape=(1, 3, self.input_dim, self.input_dim), ctx=self.ctx)
net(tmp_batch)
net.load_weights(self.params_path, fine_tune=False)
else:
raise Exception('模型错误') # 抛出异常
net.hybridize() # 编译和优化网络
return net
def download_img(img_url, out_folder, imgs_names, img_name=None):
"""
下载图片
:param img_url: 图片URL
:param out_folder: 输出文件夹
:param imgs_names: 已有图片
:param img_name: 图片名称
:return: None
"""
if not img_name:
img_name = img_url.split('/')[-1] # 图片文件名
if img_name in imgs_names:
print_info('图片已存在: %s' % img_name)
return
img_data = requests.get(img_url).content
out_file = os.path.join(out_folder, img_name) # 输出文件
with open(out_file, 'wb') as hl:
hl.write(img_data)
print_info('图片已下载: %s' % img_name)
def download_imgs(img_file, out_folder, prefix=None):
"""
下载图片集合
:param img_file: 图片文件
:param out_folder: 文件夹
:param prefix: 文件前缀
:return: None
"""
paths_list = read_file(img_file)
print_info('图片总数: %s' % len(paths_list))
_, imgs_names = traverse_dir_files(out_folder)
count = 0
for (index, path) in enumerate(paths_list):
if prefix:
download_img(path, out_folder, imgs_names,
prefix + '_' + str(index) + '.jpg')
else:
download_img(path, out_folder, imgs_names)
count += 1
if count % 200 == 0:
print_info('已下载: %s' % count)
def iou_of_boxes(boxes1, boxes2):
res_list = []
for box1 in boxes1:
final_iou = 0
final_box = None
for box2 in boxes2:
iou = bb_intersection_over_union(box1, box2)
if iou > final_iou:
final_iou = iou
final_box = box2
if final_iou > 0.5:
res_list.append((box1, final_box, final_iou))
t1, t2, tr = len(boxes1), len(boxes2), len(res_list)
if tr == 0: # 没有检测源, 则认为正确
return [1.0, 1.0]
recall = safe_div(tr, t1) # 召回率
precision = safe_div(tr, t2) # 精准率
print_info('精准: {:.4f}, 召回: {:.4f}, 匹配结果: {}'.format(
precision, recall, res_list))
return [precision, recall]
def detect_image_facets(self, image):
"""
检测图片的数据细节，检测框、置信度、类别
:param image: PIL图片
:return: 检测框、置信度、类别
"""
start_t = timer() # 起始时间
print_info('{}检测开始{}'.format('-' * 10, '-' * 10))
if self.model_image_size != (
None, None): # 416x416, 416=32*13，必须为32的倍数，最小尺度是除以32
assert self.model_image_size[
0] % 32 == 0, 'Multiples of 32 required'
assert self.model_image_size[
1] % 32 == 0, 'Multiples of 32 required'
boxed_image = letterbox_image(image,
tuple(reversed(
self.model_image_size))) # 填充图像
else:
new_image_size = (image.width - (image.width % 32),
image.height - (image.height % 32))
boxed_image = letterbox_image(image, new_image_size) # 强制转换为32的倍数
image_data = np.array(boxed_image, dtype='float32') # 图片数据
image_data /= 255. # 转换0~1
image_data = np.expand_dims(image_data, 0) # 添加批次维度，将图片增加1维
# 参数盒子、得分、类别；输入图像0~1，4维；原始图像的尺寸
out_boxes, out_scores, out_classes = self.sess.run(
[self.boxes, self.scores, self.classes],
feed_dict={
self.yolo_model.input: image_data, # 输入检测图片
self.input_image_shape: [image.size[1],
image.size[0]], # 输入检测尺寸
K.learning_phase(): 0 # 学习率, 0表示测试, 1表示训练
})
print_info('检测时间: {:.4f} 秒'.format(timer() - start_t)) # 检测执行时间
print_info('检测物体数: {} 个'.format(len(out_boxes))) # 检测结果
print_info('{}检测结束{}'.format('-' * 10, '-' * 10))
return out_boxes, out_scores, out_classes
def verify_model(self):
"""
处理标记文件夹
:param img_folder: 图片文件夹
:param out_folder:
:return:
"""
img_dict = format_img_and_anno(self.img_folder)
res_list = []
for count, img_name in enumerate(img_dict):
print_info('-' * 50)
print_info('图片: {}'.format(img_name))
(img_p, anno_p) = img_dict[img_name]
try:
res_dict = self.detect_img(img_p, anno_p, self.out_folder)
except Exception as e:
print_ex('检测异常: {}'.format(e))
continue
res_list.append(res_dict)
# if count == 10:
# break
print_info('-' * 50)
for target_name in (['all'] + self.targets_name):
if target_name in ['truck', 'bus']:
continue
ap, ar, count = 0, 0, 0
for pr_dict in res_list:
if target_name in pr_dict:
tp, tr = pr_dict[target_name]
ap += tp
ar += tr
count += 1
mAp = safe_div(ap, count)
mAr = safe_div(ar, count)
print_info('类: {} P: {:.4f} %, R: {:.4f} %'.format(
target_name, mAp * 100, mAr * 100))
def process_anno_folder(img_folder, out_folder):
img_dict = format_img_and_anno(img_folder)
_, file_names = traverse_dir_files(out_folder)
for count, img_name in enumerate(img_dict.keys()):
print_info('-' * 50)
(img_p, anno_p) = img_dict[img_name]
if not img_p or not anno_p:
print_info('图片: {} 异常'.format(img_name))
continue
if img_name in file_names:
print_info('图片: {} 已存在'.format(img_name))
else:
print_info('图片: {}'.format(img_name))
image_data = Image.open(img_p)
boxes_list, name_list = read_anno_xml(anno_p)
prob_list = [1.0 for x in range(len(boxes_list))]
img_boxes = draw_boxes_simple(image_data, boxes_list, prob_list,
name_list)
img_boxes.save(os.path.join(out_folder, img_name + '.d.jpg'))
def main():
ym = Y3Model()
img_folder = os.path.join(IMG_DATA, 'jiaotong-0727')
right_folder = os.path.join(IMG_DATA, 'jiaotong-0727-right')
wrong_folder = os.path.join(IMG_DATA, 'jiaotong-0727-wrong')
none_folder = os.path.join(IMG_DATA, 'jiaotong-0727-none')
mkdir_if_not_exist(right_folder, is_delete=True)
mkdir_if_not_exist(wrong_folder, is_delete=True)
mkdir_if_not_exist(none_folder, is_delete=True)
img_dict = format_img_and_anno(img_folder)
r_count = 0
all_count = 0
no_recall_count = 0
for count, img_name in enumerate(img_dict):
(img_p, anno_p) = img_dict[img_name]
# print(img_p)
try:
tag_res, img_box = ym.detect_img(img_p, True)
except Exception as e:
continue
w_tags = []
for tag in tag_res.keys():
if tag_res[tag] <= 0.01:
print_info('删除Tag {} {}'.format(tag, tag_res[tag]))
w_tags.append(tag)
for tag in w_tags:
tag_res.pop(tag, None) # 小于1%的类别
all_count += 1
p_classes = set(tag_res.keys())
_, t_classes = read_anno_xml(anno_p)
merge_dict = {'truck': 'car', 'bus': 'car', 'car': 'car'} # 合并类别
t_classes = map_classes(merge_dict, t_classes) # 合并类别
traffic_names = [
'bicycle', 'car', 'motorbike', 'aeroplane', 'bus', 'train',
'truck', 'boat'
]
t_classes = set(t_classes) & set(traffic_names)
img_name = img_p.split('/')[-1]
is_right = False
if p_classes and p_classes.issubset(t_classes): # 检测正确
r_count += 1
img_box.save(os.path.join(right_folder, img_name + '.d.jpg'))
is_right = True
elif not p_classes and not t_classes: # 空，检测正确
r_count += 1
if not img_box:
img_box = Image.open(img_p)
img_box.save(os.path.join(right_folder, img_name + '.d.jpg'))
is_right = True
elif not p_classes and t_classes: # 检测为空，实际有类
if not img_box:
img_box = Image.open(img_p)
img_box.save(os.path.join(none_folder, img_name + '.d.jpg'))
no_recall_count += 1 # 未召回
r_count += 1
is_right = True
else: # 其他，检测错误
if not img_box:
img_box = Image.open(img_p)
img_box.save(os.path.join(wrong_folder, img_name + '.d.jpg'))
print_info('P: {}, T: {}, {}'.format(list(p_classes), list(t_classes),
'正确' if is_right else '错误'))
right_ratio = safe_div(r_count, all_count)
print_info('正确: {}, 全部: {}, 未召回: {}, 准确率: {}'.format(
r_count, all_count, no_recall_count, right_ratio))
def detect_img(self, img_path, xml_path, out_folder=None):
color_list_1 = make_line_colors(n_color=20, alpha=0.6, bias=1.0)
color_list_2 = make_line_colors(n_color=20, alpha=1.0, bias=0.8)
car_list = ['truck', 'bus', 'car']
# img_path = os.path.join(IMG_DATA, 'jiaotong-0727', '6emekgFq0neQv8ELXmCq5aQnL3Zz.jpg')
# xml_path = os.path.join(IMG_DATA, 'jiaotong-0727', '6emekgFq0neQv8ELXmCq5aQnL3Zz.xml')
img_data = Image.open(img_path)
boxes, scores, classes_no = self.detect_image_facets(img_path) # 检测图片
boxes = self.reform_boxes(boxes)
classes = [self.classes_name[int(i)]
for i in classes_no] # 将classes的no转换为name
boxes, scores, classes = filter_sbox(
(img_data.size[0], img_data.size[1]), (boxes, scores, classes))
boxes, scores, classes = self.keep_classes(self.targets_name, boxes,
scores, classes)
classes = self.merge_classes_name(car_list, 'car', classes)
print_info('检测数: {} - {}'.format(len(boxes), classes))
t_boxes, t_classes = read_anno_xml(xml_path)
t_scores = ['T' for i in range(len(t_boxes))]
t_boxes, t_scores, t_classes = \
filter_sbox((img_data.size[0], img_data.size[1]), (t_boxes, t_scores, t_classes))
t_boxes, t_scores, t_classes = self.keep_classes(
self.targets_name, t_boxes, t_scores, t_classes)
t_classes = self.merge_classes_name(car_list, 'car', t_classes)
print_info('真值数: {}'.format(len(t_boxes), t_classes))
uni_classes = sorted(list(set(classes) | set(t_classes)))
color_dict_1 = dict(zip(uni_classes, color_list_1[:len(uni_classes)]))
color_dict_2 = dict(zip(uni_classes, color_list_2[:len(uni_classes)]))
img_true = draw_boxes_simple(img_data, t_boxes, t_scores, t_classes,
color_dict_2)
# img_true.show()
img_pred = draw_boxes_simple(img_data,
boxes,
scores,
classes,
color_dict_1,
is_alpha=True)
# img_pred.show()
if out_folder:
img_name = img_path.split('/')[-1]
out_img = os.path.join(out_folder, img_name + '.d.png')
img_pred.save(out_img) # 存储
res_dict = dict()
# print('all')
print_info('交通工具')
ap, ar = self.iou_of_boxes(t_boxes, boxes)
res_dict['all'] = (ap, ar)
print_info('')
for class_name in uni_classes:
print_info('类别: {}'.format(class_name))
sub_t_boxes, sub_t_classes = self.filter_class(
class_name, t_boxes, t_classes)
sub_boxes, sub_classes = self.filter_class(class_name, boxes,
classes)
ap, ar = self.iou_of_boxes(sub_t_boxes, sub_boxes)
res_dict[class_name] = (ap, ar)
return res_dict