MOT：MOTchallenge任务评价方法

川长思鸟来 2022-09-01 14:55 11阅读 0赞

## GT介绍 ##

1,1,912,484,97,109,0,7,1
    2,1,912,484,97,109,0,7,1
    3,1,912,484,97,109,0,7,1
    4,1,912,484,97,109,0,7,1
    5,1,912,484,97,109,0,7,1
    6,1,912,484,97,109,0,7,1
    7,1,912,484,97,109,0,7,1

<table> 
 <thead> 
  <tr> 
   <th>index</th> 
   <th>0</th> 
   <th>1</th> 
   <th>2</th> 
   <th>3</th> 
   <th>4</th> 
   <th>5</th> 
   <th>6</th> 
   <th>7</th> 
   <th>8</th> 
  </tr> 
 </thead> 
 <tbody> 
  <tr> 
   <td>value</td> 
   <td>1</td> 
   <td>1</td> 
   <td>912</td> 
   <td>484</td> 
   <td>97</td> 
   <td>109</td> 
   <td>0</td> 
   <td>7</td> 
   <td>1</td> 
  </tr> 
  <tr> 
   <td>含义</td> 
   <td>frame id</td> 
   <td>track id</td> 
   <td>x</td> 
   <td>y</td> 
   <td>w</td> 
   <td>h</td> 
   <td>mark</td> 
   <td>class</td> 
   <td>vis ratio</td> 
  </tr> 
 </tbody> 
</table>

*  mark标记为0不可用，1为可用
 *  class有12个类别
 *  vis ratio为可见部分比例，最大为1，最小为0

<table> 
 <thead> 
  <tr> 
   <th>ID</th> 
   <th>label</th> 
  </tr> 
 </thead> 
 <tbody> 
  <tr> 
   <td>1</td> 
   <td>Pedestrian</td> 
  </tr> 
  <tr> 
   <td>2</td> 
   <td>Person on vehicle</td> 
  </tr> 
  <tr> 
   <td>3</td> 
   <td>Car</td> 
  </tr> 
  <tr> 
   <td>4</td> 
   <td>Bicycle</td> 
  </tr> 
  <tr> 
   <td>5</td> 
   <td>Motorbike</td> 
  </tr> 
  <tr> 
   <td>6</td> 
   <td>Non motorized vechile</td> 
  </tr> 
  <tr> 
   <td>7</td> 
   <td>Static person</td> 
  </tr> 
  <tr> 
   <td>8</td> 
   <td>Distractor</td> 
  </tr> 
  <tr> 
   <td>9</td> 
   <td>Occluder</td> 
  </tr> 
  <tr> 
   <td>10</td> 
   <td>Occluder on the ground</td> 
  </tr> 
  <tr> 
   <td>11</td> 
   <td>Occluder full</td> 
  </tr> 
  <tr> 
   <td>12</td> 
   <td>Reflection</td> 
  </tr> 
 </tbody> 
</table>

## 评价方法 ##

1.  定义实际用于指标计算的valid bbox，label id == 1且 mark==1；
2.  定义需要忽略的ignore bbox，label id == 2，7，8，12 或者vis ratio==0；
3.  将预测结果首先与ignore bbox进行匹配，将于ignore bbox匹配上的预测值剔除，即这些值虽然不在gt里，但是也不算tp；
4.  将剩余的预测结果，与valid bbox进行匹配，得到最终的结果。

## Reference： ##

*  [py-motmetrics][]
 *  [TrackEval][]

[py-motmetrics]: https://github.com/cheind/py-motmetrics
[TrackEval]: https://github.com/JonathonLuiten/TrackEval