search

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
250 stars 21 forks source link

Reproducing Table 2 #41

Closed helen1c closed 2 months ago

helen1c commented 4 months ago

Hi @w1oves,

Congratulations on the excellent work.

I am keen to reproduce Table 2 from your paper but have been unable to locate all the necessary configuration files. Could you please provide guidance on which scripts to run and the specific hyperparameters required? I am particularly interested in the "Freeze" and "Rein" rows.

Thank you in advance for your assistance.

w1oves commented 4 months ago

I will be releasing these checkpoints. Given the substantial number of checkpoints to publish, I will prioritize the release of those related to the dinov2 and clip models. It's worth noting that the checkpoint for dinov2 combined with reins has already been released. Additionally, the configurations for both frozen dinov2 and the full dinov2 have been made available.

helen1c commented 4 months ago

@w1oves,

Thank you for your reply.

I am specifically interested in the configurations (not checkpoints) for all five "freeze" models (DINOV2, CLIP, MAE, SAM, and EMA) shown in Table 5. I would like to reproduce these results on my own.

Could you let me know when you expect to publish these configurations?

Thank you!

w1oves commented 4 months ago

You can find frozen dinov2 and clip config in configs/frozen_vfms. Others will be released soon.

w1oves commented 4 months ago

The configs for frozen SAM-H and EVA02-L have uploaded.