search

vqdang / hover_net

Simultaneous Nuclear Instance Segmentation and Classification in H&E Histology Images.
MIT License
522 stars 220 forks source link

How to use Mask-RCNN Classified nuclei #166

Open xiaoshuomin opened 2 years ago

xiaoshuomin commented 2 years ago

Mask-RCNN is ofen used in detection and segmentation，but it is rarely used for classification.please tell me how to modify the model so that Mask-RCNN can be used for classification tasks?Thanks

simongraham commented 2 years ago

Hi. I am not sure why you are asking about Mask-RCNN, as this is the HoVer-Net repo, but I can answer your question anyway.

In fact Mask-RCNN is purposely built for multi-class segmentation. Have a look at the original paper here and you will see that. The method will produce a set of predicted bounding boxes and then for each box, it will predict the class and produce the segmentation mask. Class prediction is done using a series of fully connected layers and so can easily be adapted to produce as many classes as desired.

xiaoshuomin commented 2 years ago

I am trying to reproduce the results of Mask-RCNN in the paper. I think I know what to do. Thank you for your answer

simongraham commented 2 years ago

Please also take a look at issue #40

xiaoshuomin commented 2 years ago

Hi, I am reproducing your work in your paper.Thank you for your previous answer,but I still have some questions.Instance segmentation for the nuclear type classification tasks use the notion of distance(6 pixels at 20 × or 12 pixels at 40 × ) to determine whether nuclei have been detected. What indicator is used for the detection method of nuclear classification task to determine whether the nucleus has been detected?