数据集格式转换一:DarkLabelToMOT
1. 背景
使用DarkLabel
制作目标追踪数据集,在训练模型时需要用到MOT数据集
格式,因此需要进行格式转换
。
2. 参考
参考pprpp大神的文章:https://www.jianshu.com/p/f36a45236044
先说区别:时间太长,我已经忘记了许多,似乎记得pprpp大神的代码应该是多个文件夹下的自动转换,而这里的代码是手动进行每个子文件夹的转换。pprpp大神的应该是单类别转换,这里的是多类别转换
。
3. 单个文件夹的gt文件转换
import os
'''
(1)区别如下:
dir_dl_gt.txt(转换前)
frame(从0开始计), 数量, id(从0开始), box(x1,y1,x2,y2), class=null
0,4,0,450,194,558,276,null,1,408,147,469,206,null,2,374,199,435,307,null,3,153,213,218,314,null
1,4,0,450,194,558,276,null,1,408,147,469,206,null,2,374,199,435,307,null,3,153,213,218,314,null
2,4,0,450,194,558,276,null,1,408,147,469,206,null,2,374,199,435,307,null,3,153,213,218,314,null
-----------------------------------------------------------------------------------------------
gt.txt(转换后)
frame(从1开始计), id(从1开始计), box(left top w, h),ignore=1(不忽略), class=1(从1开始),覆盖=1),
1,1,1363,569,103,241,1,1,0.86014
2,1,1362,568,103,241,1,1,0.86173
3,1,1362,568,103,241,1,1,0.86173
4,1,1362,568,103,241,1,1,0.86173
(2)类别信息:
class1,class2,class3,class4,class5为追踪目标的类别
'''
def xyxy2xywh(x):
# Convert bounding box format from [x1, y1, x2, y2] to [x, y, w, h]
# y = torch.zeros_like(x) if isinstance(x,
# torch.Tensor) else np.zeros_like(x)
y = [0, 0, 0, 0]
y[0] = (x[0] + x[2]) / 2
y[1] = (x[1] + x[3]) / 2
y[2] = x[2] - x[0]
y[3] = x[3] - x[1]
return y
def process_darklabel(video_label_path, mot_label_path):
f = open(video_label_path, "r")
f_o = open(mot_label_path, "w")
contents = f.readlines()
for line in contents:
line = line[:-1]
num_list = [num for num in line.split(',')]
frame_id = int(num_list[0]) + 1 # 帧id
total_num = int(num_list[1]) # 图片中的目标个数
base = 2 # 从2开始
for i in range(total_num): # 多个目标
_id = int(num_list[base + i * 6]) + 1
_box_x1 = int(num_list[base + i * 6 + 1])
_box_y1 = int(num_list[base + i * 6 + 2])
_box_x2 = int(num_list[base + i * 6 + 3])
_box_y2 = int(num_list[base + i * 6 + 4])
y = xyxy2xywh([_box_x1, _box_y1, _box_x2, _box_y2])
classname_str = num_list[base + i * 6 + 5] # 多类别多目标追踪,帧id和fairmot一样从1开始,类别id从0开始
if(classname_str=="class1"):
class_index = 0
elif(classname_str=="class2"):
class_index = 1
elif(classname_str=="class3"):
class_index = 2
elif (classname_str == "class4"):
class_index = 3
elif (classname_str == "class5"):
class_index = 4
else:
class_index = 5
print('warning:文件中存在未识别的类别信息——',classname_str)
write_line = "%d,%d,%d,%d,%d,%d,1,%d,1\n" % (frame_id, _id, y[0], y[1], y[2], y[3], int(class_index))
f_o.write(write_line)
f.close()
f_o.close()
if __name__ == "__main__":
root_dir = "./path/work/" # 图片文件所在文件夹
video_label_path = os.path.join(root_dir, "VID10_dl_gt.txt") # darklabel标注保存文件
mot_label_path = os.path.join(root_dir, "gt.txt") # 生成的mot标注文件
process_darklabel(video_label_path, mot_label_path)
4. 转换效果截图
5. 按照MOT数据集文件格式进行整理
将转换得到的gt.txt文件放到gt文件夹中;
img1中方放图片,记得删除dir_dl_gt.txt文件(darklabel标注文件);
seqinfo.ini文件保存的是该文件夹的信息,各种信息按照个人理解如下图所示;
det文件夹先不用管,我一般将darklabel标注文件放到这里,具体的MOT数据集格式讲解可自行百度。
PS:鼠标坏了总是连击,好几次截图在编辑过程中就跟确认了(QQ截图,双击完成),没有好好排版,各位别嫌弃丑啊 -_-| 。