l1 = np.mean(np.abs(t))
real_center = (real_center + 1) * 127.5
fake = (fake + 1) * 127.5
def psnr(img1, img2):
mse = np.mean((img1 - img2)**2)
if mse == 0:
return 100
PIXEL_MAX = 255.0
return 20 * math.log10(PIXEL_MAX / math.sqrt(mse))
for i in range(1):
p = p + psnr(real_center[i].transpose(1, 2, 0), fake[i].transpose(
1, 2, 0))
ssim_none = ssim(out_crop, target_crop)
sum_l1 = sum_l1 + l1
sum_l2 = sum_l2 + l2
sum_psnr_2 = sum_psnr_2 + p
sum_ssim_2 = sum_ssim_2 + ssim_none
outstr = "[Global-local] [%4d]-> " % count
outstr = outstr + "l1:\t%f\t" % l1
outstr = outstr + "l2:\t%f\t" % l2
outstr = outstr + "psnr:\t%f\t" % p
outstr = outstr + "ssim_img :\t%f\t \n" % ssim_none
print(outstr)
f.write(outstr)
outresultstr = "[Global-local] Total-> psnr:\t%f\t" % (sum_psnr_2 / count)
outresultstr = outresultstr + "ssim_img :\t%f\t \n" % (sum_ssim_2 / count)
for batch in testing_data_loader:
input, target, input_masked = Variable(batch[0], volatile=True), Variable(batch[1], volatile=True), Variable(
batch[2], volatile=True)
if opt.cuda:
netG = netG.cuda()
input_masked = input_masked.cuda()
target = target.cuda()
out = netG(input_masked)
count = count + 1
mse = criterionMSE(out, target)
psnr = 10 * math.log10(1 / mse.data[0])
ssim_none = ssim(out, target)
sum_psnr = sum_psnr + psnr
sum_ssim = sum_ssim + ssim_none
outstr = "[%4d]-> " % count
outstr = outstr + "psnr:\t%f\t" % psnr
outstr = outstr + "ssim_img :\t%f\t \n" % ssim_none
print(outstr)
f.write(outstr)
out = out.cpu()
out_img = out.data[0]
def ToPilImage(image_tensor):
def torch_ssim(img, ref): #input [28,256,256]
return ssim(torch.unsqueeze(img, 0), torch.unsqueeze(ref, 0))
############## Stage_1 ##############
if opt.cuda:
netG_1st = netG_1st.cuda()
input_masked = input_masked.cuda()
target = target.cuda()
input_2ndmasked =input_2ndmasked.cuda()
input_small = down_sample(input)
target_small = down_sample(target)
out_1st = netG_1st(down_sample(input_masked))
out_1st_up = up_sample(out_1st)
mse = criterionMSE(out_1st, target_small)
psnr = 10 * math.log10(1 / mse.data[0])
ssim_none = ssim(out_1st, target_small)
sum_psnr_1 = sum_psnr_1 + psnr
sum_ssim_1 = sum_ssim_1 + ssim_none
outstr = "[stage 1] [%4d]-> " % count
outstr = outstr + "psnr:\t%f\t" % psnr
outstr = outstr + "ssim_img :\t%f\t \n" % ssim_none
print(outstr)
f.write(outstr)
input_small = input_small.cpu().data[0]
out_1st = out_1st.cpu().data[0]
target_small = target_small.cpu().data[0]
merged_result_1st = torch.cat((input_small, out_1st, target_small), 1)
#save_img(merged_result_1st, "result/{}/{}_{}_{}.jpg".format(opt.dataset, opt.model_1st, count, opt.dataset))