search

tnmichael309 / kaggle-nuclei-segmentation

Kaggle Nuclei Segmentation (top 4% solution)
11 stars 1 forks source link

Question about running code of deform_conv part #1

Closed satchmo233 closed 5 years ago

satchmo233 commented 5 years ago

I just want to know did you run the 'deform_conv' part successfully? how can i get off keras,and embedding the deform_conv layer in the resnet101?

tnmichael309 commented 5 years ago

If you don't want to use it, you could simply use mask rcnn from model.py instead of model_deform_conv.py (model.py is where most other codes developed: https://github.com/tnmichael309/kaggle-nuclei-segmentation/blob/master/Mask_RCNN/model.py)

To experiment with deform:

import model_deform_conv as modellib

see identity_block\conv_block\resnet_graph in: https://github.com/tnmichael309/kaggle-nuclei-segmentation/blob/master/Mask_RCNN/model_deform_conv.py

which uses ConvOffset2D implementation in : https://github.com/tnmichael309/kaggle-nuclei-segmentation/blob/master/Mask_RCNN/deform_conv.py

satchmo233 commented 5 years ago

Thanks for you help! Now I have knew how to change resnet and embed the deform_conv layer in it,after research you code. I really appreciate you very much.