search

cv516Buaa / ST-DASegNet

Apache License 2.0
25 stars 4 forks source link

Question about Comparative experiments #8

Open jetitime opened 1 year ago

jetitime commented 1 year ago

Dear Author, Sorry to distrub you again. In your paper, you mentioned that you reproduced two DAFormer-based domain adaptation experiments from Potsdam to Vaihingen and from Vaihingen to Potsdam. I cut the picture to the size of 512x512 according to your settings, and it is strange that the experimental result miou from Potsdam to Vaihingen is 61.97%, which is comparable to your result, but the experimental result from Vaihingen to Potsdam miou is 62.89%, which is far more than your reproduction result.I just modified in config/base/datasets/uda_isprs_to_potsdam_512x512.py. Can you give some advise about is? I have no idea why the result I reproduction is so strange. Looking forward to your reply. Thank you!

cv516Buaa commented 1 year ago

I am really glad that you can reproduce a competitive results on DAFormer. Actually, I do not reproduce such a high results. Maybe flunctuation occurs between different runs. Therefore, I think that you can conduct different runs. Thanks for continuously following our paper and code!!

jetitime commented 1 year ago

Thanks for your reply. I'll try again.

jetitime commented 1 year ago

Sorry to distrub you again. Can you provide DAFormer-based domain adaptation logs from vaihingen to potsdam?I want to find the problem of my own experiment through comparison. In addition, I would like to ask if the mitb5 weight file you used to reproduce the DAFormer experiment was officially provided by segformer or by you in st-dasegnet? I found that the two seemed to be different. Looking forward to your reply. Thank you! 20230929_095531.log

XZhang878 commented 9 months ago

@jetitime, I'm glad you have reproduced the remote sensing semantic segmentation on DAFormer. I have been using DAFormer to reproduce remote sensing semantic segmentation recently, but I encountered problems during the reproduction process and the reproduced results did not meet your expectations. May I ask if you can help me solve this problem? Thank you.

SSSilvester commented 4 months ago

Thanks for your reply. I'll try again.

HELLO! I've been working on reproducing the research from the paper we're both interested in and have encountered some challenges along the way. Would it be possible to exchange contact information so we can discuss these issues and collaborate? Your insights would be greatly appreciated.