w1oves / Rein

[CVPR 2024] Official implement of <Stronger, Fewer, & Superior: Harnessing Vision Foundation Models for Domain Generalized Semantic Segmentation>
https://zxwei.site/rein
GNU General Public License v3.0
215 stars 19 forks source link

Question about source domain performance. #38

Closed weeknan closed 3 months ago

weeknan commented 3 months ago

Hi,

Thank you for your insightful work, which has provided me with valuable perspectives.

I have a question regarding the performance on the source domain, specifically the mIoU on GTAV as presented in Table 4. On the source domain, does Rein outperform other methods such as VPT and Adaptformer?

Best regards.

w1oves commented 3 months ago

I haven't recorded the performance for the source domain, but based on our experiments with Cityscapes, the performance of my method should be similar to that of full parameter fine-tuning.

weeknan commented 3 months ago

ok,thx!

w1oves commented 3 months ago

When trained on cityscapes, the evaluation mIoU of Rein is 84.2, and full fine-tune is 84.2, too.