search

Francis0625 / Omni-SR

[CVPR2023] Implementation of ''Omni Aggregation Networks for Lightweight Image Super-Resolution".
192 stars 12 forks source link

Initial learning rate #11

Closed jlu-lx closed 1 year ago

jlu-lx commented 1 year ago

The learning rate in train_OmniSR_X2_DF2K.yaml is 5e-4. But in your Omni paper, the learning rate is set to 1e-4. Please tell me the reason.

Francis0625 commented 1 year ago

Sorry for the misalignment, and we will update the paper and the code. Using the provied training setting, you can achieve better results than the reported results in the paper.