search

zlckanata / DeepGlobe-Road-Extraction-Challenge

D-LinkNet: LinkNet with Pretrained Encoder and Dilated Convolution for High Resolution Satellite Imagery Road Extraction
http://openaccess.thecvf.com/content_cvpr_2018_workshops/papers/w4/Zhou_D-LinkNet_LinkNet_With_CVPR_2018_paper.pdf
MIT License
641 stars 195 forks source link

Multi class segmentation #29

Closed aihyper11 closed 3 years ago

aihyper11 commented 3 years ago

What changes are requited in the current script to use for multi class problem?

zlckanata commented 3 years ago

This repo is rough and only suitable for binary seg now, I guess it is necessary to edit several files for multi-class problem.

Dataloader(label generation) https://github.com/zlkanata/DeepGlobe-Road-Extraction-Challenge/blob/master/data.py#L91

Loss Function https://github.com/zlkanata/DeepGlobe-Road-Extraction-Challenge/blob/mastre/loss.py#L7

Network's Last Layer https://github.com/zlkanata/DeepGlobe-Road-Extraction-Challenge/blob/master/framework.py#L10 self.net = net().cuda() --> self.net = net(num_class=?).cuda()

Evaluation https://github.com/zlkanata/DeepGlobe-Road-Extraction-Challenge/blob/master/test.py

aihyper11 commented 3 years ago

Thank you very much for your reply. I will go through all the suggested changes.