search

wgcban / ChangeFormer

[IGARSS'22]: A Transformer-Based Siamese Network for Change Detection
https://www.wgcban.com/research#h.e51z61ujhqim
MIT License
428 stars 57 forks source link

DSIFN dataset splits #52

Closed zaidbhat1234 closed 1 year ago

zaidbhat1234 commented 1 year ago 
          @wgcban I had another question regarding the DSIFN dataset that I downloaded from the link in your Readme:
  1. Did you use the same train/val/test split for ChangeFormer when you compare its performance with other CD models like BIT?
  2. I notice that majority of images used in test.txt are also present in train.txt (164/192) ~85%. Is this something that is an error or it was the same in previous CD works such as BIT? Is this the reason for huge difference in metrics from previous works on DSIFN?

Originally posted by @zaidbhat1234 in https://github.com/wgcban/ChangeFormer/issues/51#issuecomment-1466909895

wgcban commented 1 year ago

@zaidbhat1234 Hello,

  1. yes, if the train/val/test splits are given in the dataset like LEVIR-CD.
  2. I think there is some bug while dividing DSIFN dataset into splits. Since they do not provide specific train/val/test splits, I created my own splits. I will examine this issue and try to reproduce the results again. Thanks for pointing out the bug.
zaidbhat1234 commented 1 year ago

@wgcban Thanks for your response.

Could you please update the new results in DSIFN dataset once you resolve the bug? Could you also please update the Readme file with the updated train/val/test splits?