search

cv516Buaa / ST-DASegNet

Apache License 2.0
25 stars 4 forks source link

Question about Comparison experiments from Potsdam R-G-B to Vaihingen IR-R-G #9

Open XZhang878 opened 10 months ago

XZhang878 commented 10 months ago

Dear author, Sorry to bother you. According to your open-source code, I reproduced the domain adaptation experiment from Potsdam RGB to Vaihingen IR-R-G. I cropped the image to a size of 512x512 according to your experimental settings. The strange thing is that the experimental result miou from Potsdam RGB to Vaihingen IR-R-G is 46.7%, but the experimental result miou from Potsdam RGB to Vaihingen IR-R-G published in your article is 55.16%, which is lower than the results in your article. Can you give some advice? I don't know why the results of my replication are so weird. Looking forward to your reply. Thank you!

cv516Buaa commented 10 months ago

Thank you for following our work. One tip for you. Batchsize really matters!! At the beginning of this work, we use two 1080ti GPUs. The batchsize cannot be set too large. With the development of our lab equipment, we have two 3090 GPUs. Things become different.

XZhang878 commented 10 months ago

Thank you very much for your answer. My experimental equipment is also 3090. I have solved this problem.