《Look More Than Once:An Detector for Text of Arbitrary Shape》

不念不忘少年蓝@ 2023-08-17 16:42 114阅读 0赞

## Look More Than Once:An Detector for Text of Arbitrary Shape ##

<table> 
 <thead> 
  <tr> 
   <th style="text-align:center;">The First Column</th> 
   <th style="text-align:center;">The Second Column</th> 
  </tr> 
 </thead> 
 <tbody> 
  <tr> 
   <td style="text-align:center;">这个是错别字吗 <img src="https://img2018.cnblogs.com/blog/1433065/201909/1433065-20190915192212834-2016734703.png" alt="1433065-20190915192212834-2016734703.png"></td> 
   <td style="text-align:center;">应该是design吧</td> 
  </tr> 
  <tr> 
   <td style="text-align:center;">Iterative Refinement Module</td> 
   <td style="text-align:center;">迭代优化模块</td> 
  </tr> 
  <tr> 
   <td style="text-align:center;">LOMO</td> 
   <td style="text-align:center;">Input Image<br>DR Outputs<br>IRM Outputs<br>SEM Outputs</td> 
  </tr> 
  <tr> 
   <td style="text-align:center;">LOMO Network</td> 
   <td style="text-align:center;">Main Network<br>DirectRegressor(DR)<br>Iterative Refinement Module(IRM)<br>Shape Expression Module(SEM)</td> 
  </tr> 
  <tr> 
   <td style="text-align:center;">DR(Direct Regressor)</td> 
   <td style="text-align:center;"></td> 
  </tr> 
  <tr> 
   <td style="text-align:center;">DR（DirectRegressor）</td> 
   <td style="text-align:center;"><img src="https://img2018.cnblogs.com/blog/1433065/201909/1433065-20190914145459285-388591337.png" alt="1433065-20190914145459285-388591337.png"><img src="https://img2018.cnblogs.com/blog/1433065/201909/1433065-20190914145557651-1313257748.png" alt="1433065-20190914145557651-1313257748.png">每个像素点都能预测<br>文字或文字四边形</td> 
  </tr> 
  <tr> 
   <td style="text-align:center;">SEM(Shape Expression Module)</td> 
   <td style="text-align:center;"><img src="https://img2018.cnblogs.com/blog/1433065/201909/1433065-20190914151807735-1875776290.png" alt="1433065-20190914151807735-1875776290.png"></td> 
  </tr> 
  <tr> 
   <td style="text-align:center;">RoItransform层后的特征图 <br> 进行两次上采样</td> 
   <td style="text-align:center;"><img src="https://img2018.cnblogs.com/blog/1433065/201909/1433065-20190914152245304-1050166145.png" alt="1433065-20190914152245304-1050166145.png"></td> 
  </tr> 
  <tr> 
   <td style="text-align:center;">学习文字行的<br>三种不同属性</td> 
   <td style="text-align:center;">文字行区域(Text Region)<br>边界偏差(Border Offsets)<br>文字行中心线(Text Center Line)<img src="https://img2018.cnblogs.com/blog/1433065/201909/1433065-20190914152217329-1846560112.png" alt="1433065-20190914152217329-1846560112.png"></td> 
  </tr> 
  <tr> 
   <td style="text-align:center;">text polygon generation</td> 
   <td style="text-align:center;">文本多边形生成</td> 
  </tr> 
  <tr> 
   <td style="text-align:center;">IRM（Iterative Refinement Module）</td> 
   <td style="text-align:center;"><img src="https://img2018.cnblogs.com/blog/1433065/201909/1433065-20190914231321707-1100740619.png" alt="1433065-20190914231321707-1100740619.png"></td> 
  </tr> 
  <tr> 
   <td style="text-align:center;">IRM输出机制<img src="https://img2018.cnblogs.com/blog/1433065/201909/1433065-20190914231847206-907790120.png" alt="1433065-20190914231847206-907790120.png"></td> 
   <td style="text-align:center;"><img src="https://img2018.cnblogs.com/blog/1433065/201909/1433065-20190914231923485-252239737.png" alt="1433065-20190914231923485-252239737.png"></td> 
  </tr> 
  <tr> 
   <td style="text-align:center;">ROItransform层之后的特征图</td> 
   <td style="text-align:center;"><img src="https://img2018.cnblogs.com/blog/1433065/201909/1433065-20190914232200652-1886947678.png" alt="1433065-20190914232200652-1886947678.png"></td> 
  </tr> 
 </tbody> 
</table>

转载于:https://www.cnblogs.com/hugeng007/p/11482614.html