jt827859032/DRRN-pytorch

Arguments for Training Command

Opened this issue · 3 comments

Ding-Liu commented

Hello! May I know the value of arguments that are used in the training command:
main.py [-h] [--batchSize BATCHSIZE] [--nEpochs NEPOCHS] [--lr LR]
[--step STEP] [--cuda] [--resume RESUME]
[--start-epoch START_EPOCH] [--clip CLIP] [--threads THREADS]
[--momentum MOMENTUM] [--weight-decay WEIGHT_DECAY]
[--pretrained PRETRAINED]

to get the PSNRs in the following table.

Scale DRRN_B1U25 Paper DRRN_B1U25 PyTorch
x2 37.74 37.69
x3 34.03 34.02
x4 31.68 31.70

For example, the values of 'nEpochs', 'step', 'batchSize' and so on...
Many thanks in advance!

jt827859032 commented

All parameters are set as default values. While during the training phase, I manually adjusted the step value to facilitate the training phase. :）

Ding-Liu commented

Thanks so much for your reply!
I used your code trying to reproduce the result, but my model is ~0.2dB lower than your model in terms of performance.

  1. Do you mean step is set as 5 when you did your training?
  2. It seems the training datafile has the name: data/train_291_32_x234.h5, which is slightly different from the name generated by the given Matlab code data/train_291_31_x234.h5. Does it means the size_input is 32 rather than 31 when your training data is generated?
jt827859032 commented

Hi, @Ding-Liu

  1. The initial step is 5 and during training in my experiment. While this value will be enlarged during training. I manually did this in the training phase and you can also change the adjustable learning rate policy here.
  2. Yeah, the h5 file contains the pathes with the size of 32 x 32.