search

minar09 / cp-vton-plus

Official implementation for "CP-VTON+: Clothing Shape and Texture Preserving Image-Based Virtual Try-On", CVPRW 2020
https://minar09.github.io/cpvtonplus/
MIT License
343 stars 120 forks source link

LR scheduling is used improperly #69

Open Erroler opened 3 years ago

Erroler commented 3 years ago

In the paper, the authors claim The learning rate was first fixed at 0.0001 for 100K steps and then linearly decays to zero for the remaining steps. However, the step method of the LR schedulers (declared on 1 2) is never called, leading to a constant learning rate throughout training. I believe the fix is to add scheduler.step() on lines 105 and 172 of train.py (PyTorch documentation regarding LR schedulers can be found here). If ok by the authors I will create a PR addressing this issue.

Erroler commented 3 years ago

On a closer look, this issue also affects the CP-VTON repository.

minar09 commented 3 years ago

Hi @Erroler , thank you very much for pointing it out. Sure, please go ahead and submit a PR. Thank you.