MOT数据集格式简介
MOT15数据集下载:https://pan.baidu.com/s/1foGrBXvsanW8BI4eybqfWg?pwd=8888
以下为一行gt示例:
1,1,1367,393,73,225,1,-1,-1,-1
各列数据对应含义如下
<frame>,<id>,<bb_left>,<bb_top>,<bb_width>,<bb_height>,<conf>,<x>,<y>,<z>
- frame:图片帧id
- id:目标id
- bb_left:bbox左上角坐标x
- bb_top:bbox左上角坐标y
- bb_width:bbox的宽度
- bb_height:bbox的高度
- conf:置信度
- x:三维坐标系x值,对于二维任务填充为-1
- y:三维坐标系y值,对于二维任务填充为-1
- z:三维坐标系z值,对于二维任务填充为-1
gt可视化
由于是跟踪任务,因此在可视化检测框的同时进一步添加箭头,用来标识目标的运动轨迹。
处理思路是读取一张图片后,同时读取两张图片的gt,若两张图片同时包含同一个目标,则用箭头连接前一帧bbox的中心点和后一帧bbox的中心点。
具体代码如下:
import cv2
def match_obj(obj_list, obj_id):
try:
index = obj_list.index(obj_id)
except:
index = -1
return index
def main(i):
img = cv2.imread("img/" + "0000{:0>2d}.jpg".format(i))
img2 = img
with open('gt/gt.txt', 'r') as f:
lines = f.readlines()
object_list = []
center_list = []
for line in lines:
img_id = line.split(',')[0]
if img_id == str(i):
object_id = line.split(',')[1]
object_list.append(object_id)
x, y, w, h = int(line.split(',')[2]), int(line.split(',')[3]), int(line.split(',')[4]), int(
line.split(',')[5])
center1 = (int(int(x) + int(w) / 2), int(int(y) + int(h) / 2))
center_list.append(center1)
if img_id == str(int(i) + 1):
object_id = line.split(',')[1]
index = match_obj(object_list, object_id)
x, y, w, h = int(line.split(',')[2]), int(line.split(',')[3]), int(line.split(',')[4]), int(
line.split(',')[5])
center2 = (int(int(x) + int(w) / 2), int(int(y) + int(h) / 2))
if index != -1:
img2 = cv2.rectangle(img, (x, y), (x + w, y + h), (0, 0, 255))
img2 = cv2.arrowedLine(img2, center_list[index], center2, (0, 255, 255), 1, 8, 0, 0.5)
cv2.imwrite("result/" + "0000{:0>2d}.jpg".format(i), img2)
if __name__ == '__main__':
for i in range(1, 52):
main(i)
可视化效果如图所示: