ALLinLLM
commented
4 years ago import torch
import time
import utility
import data
import model
import loss
from option import args
from trainer import Trainer
loader = data.Data(args)
lr_r_mean = 0
lr_g_mean = 0
lr_b_mean = 0
lr_pixels = 0
hr_r_mean = 0
hr_g_mean = 0
hr_b_mean = 0
hr_pixels = 0
loader = loader.loader_train
print(len(loader))
start = time.time()
for batch, (lr, hr, _, filename) in enumerate(loader):
lr = lr.squeeze()
hr = hr.squeeze()
lr_pixels += lr.shape[1] * lr.shape[2]
hr_pixels += hr.shape[1] * hr.shape[2]
# hr
r, g, b = lr[0], lr[1], lr[2]
lr_r_mean += torch.sum(r)
lr_g_mean += torch.sum(g)
lr_b_mean += torch.sum(b)
# lr
r, g, b = hr[0], hr[1], hr[2]
hr_r_mean += torch.sum(r)
hr_g_mean += torch.sum(g)
hr_b_mean += torch.sum(b)
print(time.time() - start)
lr_r_mean /= lr_pixels
lr_g_mean /= lr_pixels
lr_b_mean /= lr_pixels
hr_r_mean /= hr_pixels
hr_g_mean /= hr_pixels
hr_b_mean /= hr_pixels
lr_r_std = 0
lr_g_std = 0
lr_b_std = 0
lr_pixels = 0
hr_r_std = 0
hr_g_std = 0
hr_b_std = 0
hr_pixels = 0
start = time.time()
for batch, (lr, hr, _, filename) in enumerate(loader):
lr = lr.squeeze()
hr = hr.squeeze()
lr_pixels += lr.shape[1] * lr.shape[2]
hr_pixels += hr.shape[1] * hr.shape[2]
# lr
r, g, b = lr[0], lr[1], lr[2]
lr_r_std += torch.sum((r-lr_r_mean)**2)
lr_g_std += torch.sum((g-lr_g_mean)**2)
lr_b_std += torch.sum((b-lr_b_mean)**2)
# hr
r, g, b = hr[0], hr[1], hr[2]
hr_r_std += torch.sum((r-hr_r_mean)**2)
hr_g_std += torch.sum((g-hr_g_mean)**2)
hr_b_std += torch.sum((b-hr_b_mean)**2)
print(time.time() - start)
#
lr_r_std = (lr_r_std / lr_pixels) ** 0.5
lr_g_std = (lr_g_std / lr_pixels) ** 0.5
lr_b_std = (lr_b_std / lr_pixels) ** 0.5
#
hr_r_std = (hr_r_std / hr_pixels) ** 0.5
hr_g_std = (hr_g_std / hr_pixels) ** 0.5
hr_b_std = (hr_b_std / hr_pixels) ** 0.5
print('lr r,g,b mean std:', filename, lr_r_mean, lr_g_mean, lr_b_mean, lr_r_std, lr_g_std, lr_b_std)
print('hr r,g,b mean std:', filename, hr_r_mean, hr_g_mean, hr_b_mean, hr_r_std, hr_g_std, hr_b_std)
rongleiji
In #40, you said you are doing data normaliztion. but i have no idea where 127.5 come from.
as far as i know, data normalization is something like y = (x - μ)/σ, where μ is mean, and σ is the standard viarence. I guess [r, g, b]_means you used here may be the avg value for DIV2K_train_LR_bicubic:
R - 0.4488, G - 0.4371, B - 0.4040,
However, in forward, you used x = (x - self.rgb_mean.cuda()*255)/127.5 so i am confused where the σ = 127.5 come from? it should be [r, g, b]_σ but here is only one σ
And another question is which dataset to use when calculate mean RGB, HR or LRx2, x3, x4. You said in the Experimental Results of your paper that you subtracted the mean RGB of DIV2K training set. I have tested that the mean value is different between HR and LRx2,x3,x4.