search

Haochen-Wang409 / U2PL

[CVPR'22 & IJCV'24] Semi-Supervised Semantic Segmentation Using Unreliable Pseudo-Labels & Using Unreliable Pseudo-Labels for Label-Efficient Semantic Segmentation
Apache License 2.0
436 stars 61 forks source link

About the learning rate #83

Closed LWHYC closed 2 years ago

LWHYC commented 2 years ago

Hi! Thank you for the contribution! I'm running the semi experiments on 4 A100(80GB) GPU and set the batchsize on each GPU to 4. So I suppose I should set the learning rate to 0.005?

Thanks.

Haochen-Wang409 commented 2 years ago

If you plan to train a semi-supervised model, the learning rate should be set to 0.001 on VOC 2012 and 0.01 on Cityscapes, respectively, since the batch_size equals to 16 (4x4), which is the same as the batch_size in config.yaml (i.e., 4x4 for VOC 2012 and 2x8 for Cityscapes).

LWHYC commented 2 years ago

Thanks for your quick response!