Awesome
CDNet: Centripetal Direction Network for Nuclear Instance Segmentation
[ICCV2021]
The code includes training and inference procedures for CDNet.
Tips: There is a result written mistake (U-Net) in Table 4 in the original paper. The correct result is：
MoNuSeg
<table><tbody> <!-- START TABLE --> <!-- TABLE HEADER --> <th valign="bottom">Method Name</th> <th valign="bottom">Dice</th> <th valign="bottom">AJI</th> <!-- TABLE BODY --> <!-- U-Net --> <tr><td align="left">U-Net</a></td> <td align="center">0.8184</td> <td align="center">0.5910</td> </tr> <!-- Mask-RCNN --> <tr><td align="left">Mask-RCNN</a></td> <td align="center">0.7600</td> <td align="center">0.5460</td> </tr> <!-- DCAN --> <tr><td align="left">DCAN</a></td> <td align="center">0.7920</td> <td align="center">0.5250</td> </tr> <!-- Micro-Net --> <tr><td align="left">Micro-Net</a></td> <td align="center">0.7970</td> <td align="center">0.5600</td> </tr> <!-- DIST --> <tr><td align="left">DIST</a></td> <td align="center">0.7890</td> <td align="center">0.5590</td> </tr> <!-- CIA-Net --> <tr><td align="left">CIA-Net</a></td> <td align="center">0.8180</td> <td align="center">0.6200</td> </tr> <!-- FullNet --> <tr><td align="left">U-Net</a></td> <td align="center">0.8027</td> <td align="center">0.6039</td> </tr> <!-- Hover-Net --> <tr><td align="left">Hover-Net</a></td> <td align="center">0.8260</td> <td align="center">0.6180</td> </tr> <!-- BRP-Net --> <tr><td align="left">BRP-Net</a></td> <td align="center"> - </td> <td align="center">0.6422</td> </tr> <!-- PFF-Net --> <tr><td align="left">PFF-Net</a></td> <td align="center">0.8091</td> <td align="center">0.6107</td> </tr> <!-- Our CDNet --> <tr><td align="left">Our CDNet</a></td> <td align="center">0.8316</td> <td align="center">0.6331</td> </tr> </tbody></table>Getting Started
Create a data folder(/data) and put the datasets(MoNuSeg, CPM17) in it.
Train
cd CDNet/
python train.py
Test
cd CDNet/
python test.py