about the results of MegaAgeAsian

[oukohou]

I reimplement the SSRNet in pytorch, tested the test datasets, and got results as:

so seems my implementation is not so well as the paper's.
the parameters are like:

batch_size = 50
input_size = 64
num_epochs = 90
learning_rate = 0.001 # originally 0.001
weight_decay = 1e-4 # originally 1e-4
augment = False
optimizer_ft = optim.Adam(params_to_update, lr=learning_rate, weight_decay=weight_decay)
criterion = nn.MSELoss()
lr_scheduler = optim.lr_scheduler.StepLR(optimizer_ft, step_size=30, gamma=0.1)

I issed an issue here: SSR-Net #38, and @shamangary suggested a counsel here.
So here I'm writing for a detailed discussion about your training details, is there anything I missed or did wrongly?

My pytorch codes is here: SSR-Net-Pytorch