zifeiyu_tan

> I want to test model on test-dev set, but the code pose_estimation/valid.py is only for testing on validation set (5000 images): > > So, I have a question: How...

Does reducing batch_size affect the performance of the model?

[https://github.com/zhang943/lpn-pytorch/issues/1](url)

> same question! > no train.py module. In this link, the author mentions that the train.py used is code from HR-NET. https://github.com/zhang943/lpn-pytorch/issues/1

> This train retrained adjust lr, so I think HR-NET train is not necessarily match exactly. The author said hr-net train.py was used, but the LPN50 I trained in this...

Above are the results of testing on Val2017

This is my training log。 [lpn50_256x192_gd256x2_gc_2020-09-23-11-07_train.log](https://github.com/zhang943/lpn-pytorch/files/5273131/lpn50_256x192_gd256x2_gc_2020-09-23-11-07_train.log)

Thank you very much for your work.