关于YOLOV1
本文主要偏向代码实战,详细的算法原理着墨不多
主要创新:
实现了实时的目标检测
附上YOLOV1论文地址:
https://arxiv.org/abs/1506.02640
关于数据
首先要了解该网络架构的输入和输出,以及网络是如何进行训练的。这里主要讲一下数据集的格式以及使用labelme制作数据集(当然还有其他更好的工具,这里只做参考)
如果已经安装了python,使用以下命令直接安装labelme工具,没安python的先安python
bash
pip install labelme
之后在cmd中输入labelme,然后回车就能看到labelme的前端UI界面了
(ps:在启动labelme时也可以指定很多参数,比如 --flags可以启动时指定有哪些类别,具体不在赘述,感兴趣的读者自行搜索)
bash
labelme --flags F:\python\yolo系列\yolov1\flags.txt
之后OpenDir打开图片所在的文件夹路径,然后鼠标右键Create Rectangle夸夸框就完事了,之后会save为json格式的文件。(把自动保存打开的话会保存到图片的路径下如图:随便标了十五张)
如何你有仔细看过YOLOV1的论文会发现json根本不是我们训练所需要的格式,json多了很多无用的信息,甚至Rectangle也只标记了左上和右下两个点。别慌,我们用脚本转化一下:
python
import json
import os
def convert(img_size, box):
x1 = box[0]
y1 = box[1]
x2 = box[2]
y2 = box[3]
center_x = (x1 + x2) * 0.5 / img_size[0]
center_y = (y1 + y2) * 0.5 / img_size[1]
w = abs((x2 - x1)) * 1.0 / img_size[0]
h = abs((y2 - y1)) * 1.0 / img_size[1]
return (center_x, center_y, w, h)
def decode_json(jsonfloder_path, json_name):
txt_name = jsonfloder_path + json_name[0:-5] + '.txt'
txt_file = open(txt_name, 'w') # te files
json_path = os.path.join(json_folder_path, json_name)
data = json.load(open(json_path, 'r'))
img_w = data['imageWidth'] # json是一个字典的形式
img_h = data['imageHeight']
for i in data['shapes']:
if (i['shape_type'] == 'rectangle'):
x1 = int(i['points'][0][0])
y1 = int(i['points'][0][1])
x2 = int(i['points'][1][0])
y2 = int(i['points'][1][1])
if x1 < 0 or x2 < 0 or y1 < 0 or y2 < 0:
continue
else:
bb = (x1, y1, x2, y2)
bbox = convert((img_w, img_h), bb)
## 这里标签要对其
if i['label'] == "person":
txt_file.write("0 " + " ".join([str(a) for a in bbox]) + '\n')
elif i['label'] == "plane":
txt_file.write("1 " + " ".join([str(a) for a in bbox]) + '\n')
elif i['label'] == "dog":
txt_file.write("2 " + " ".join([str(a) for a in bbox]) + '\n')
elif i['label'] == "cat":
txt_file.write("3 " + " ".join([str(a) for a in bbox]) + '\n')
elif i['label'] == "car":
txt_file.write("4 " + " ".join([str(a) for a in bbox]) + '\n')
elif i['label'] == "horse":
txt_file.write("5 " + " ".join([str(a) for a in bbox]) + '\n')
elif i['label'] == "can":
txt_file.write("6 " + " ".join([str(a) for a in bbox]) + '\n')
elif i['label'] == "train":
txt_file.write("7 " + " ".join([str(a) for a in bbox]) + '\n')
if __name__ == "__main__":
json_folder_path = r'F:\python\yolo系列\yolov1\images' # json的path
json_names = os.listdir(json_folder_path) # file name
for json_name in json_names: # output all files
print(json_name)
if json_name[-5:] == '.json': # just work for json files
decode_json(json_folder_path, json_name)
不出意外的话会得到如下图所示的txt标签,这才是训练模型所需要的格式