search

cv516Buaa / ST-DASegNet

Apache License 2.0
25 stars 3 forks source link

About training command #5

Closed jetitime closed 1 year ago

jetitime commented 1 year ago

Hello, author. Can you tell me the meaning of '2' in './tools/dist_train.sh ./experiments/deeplabv3/config/ST-DASegNet_deeplabv3plus_r50-d8_4x4_512x512_40k_Potsdam2Vaihingen.py 2' ? Thank you very much!

cv516Buaa commented 1 year ago

gpu number

jetitime commented 1 year ago

I have changed the training progress by 1 gpu. Thank you very much!!

jetitime commented 1 year ago

Sorry to disturb you again! From the passage I learned how to realize domain disentangled module, but I still puzzle about principle of feature disentangled. It's sad that I didn't find useful article about feature disentangled. Is there any recommended article? Thank you very much!

cv516Buaa commented 1 year ago

In our paper, we do not refer to any paper specifically designing domain disentangled module. we may not the first to propose the concept of domain disentangling in UDA segmentation direction, but ST-DASegNet still provides a new paradigm. When designing DDM, we refer to the attention mechanism of SENet and DANet. Thanks for you Q.

jetitime commented 1 year ago

Thanks for your reply！I’ll learn about SENet and DANet.