def val(val_loader, model, epoch):
model.eval()
val_psnr = 0
for iteration, batch in enumerate(val_loader, 1):
input, HR = Variable(batch[0], volatile=True), Variable(batch[1])
if opt.cuda:
input = input.cuda()
HR = HR.cuda()
SR = model(input)
val_psnr += PSNR(SR.cpu().data[0].numpy(), HR.cpu().data[0].numpy())
val_psnr = val_psnr / len(val_loader)
print("PSNR = {:.3f}".format(val_psnr))
if opt.show:
writer.add_scalar('Val/PSNR', val_psnr, epoch)
def main():
input_path = '/media/hdisk/liqiang/hyperSR/test/' + opt.datasetName + '/' + str(
opt.upscale_factor) + '/'
out_path = '/media/hdisk/liqiang/hyperSR/result/' + opt.datasetName + '/' + str(
opt.upscale_factor) + '/' + opt.method + '/'
if not os.path.exists(out_path):
os.makedirs(out_path)
PSNRs = []
SSIMs = []
SAMs = []
if not os.path.exists(out_path):
os.makedirs(out_path)
if opt.cuda:
print("=> use gpu id: '{}'".format(opt.gpus))
os.environ["CUDA_VISIBLE_DEVICES"] = opt.gpus
if not torch.cuda.is_available():
raise Exception(
"No GPU found or Wrong gpu id, please run without --cuda")
model = SFCSR(opt)
if opt.cuda:
model = nn.DataParallel(model).cuda()
checkpoint = torch.load(opt.model_name)
model.load_state_dict(checkpoint['model'])
model.eval()
images_name = [x for x in listdir(input_path) if is_image_file(x)]
T = 0
for index in range(len(images_name)):
mat = scio.loadmat(input_path + images_name[index])
hyperLR = mat['LR'].astype(np.float32).transpose(2, 0, 1)
HR = mat['HR'].astype(np.float32).transpose(2, 0, 1)
input = Variable(torch.from_numpy(hyperLR).float(),
volatile=True).contiguous().view(
1, -1, hyperLR.shape[1], hyperLR.shape[2])
SR = np.array(HR).astype(np.float32)
if opt.cuda:
input = input.cuda()
localFeats = []
start = time.time()
for i in range(input.shape[1]):
if i == 0:
x = input[:, 0:3, :, :]
y = input[:, 0, :, :]
elif i == input.shape[1] - 1:
x = input[:, i - 2:i + 1, :, :]
y = input[:, i, :, :]
else:
x = input[:, i - 1:i + 2, :, :]
y = input[:, i, :, :]
output, localFeats = model(x, y, localFeats, i)
SR[i, :, :] = output.cpu().data[0].numpy()
end = time.time()
print(end - start)
T = T + (end - start)
SR[SR < 0] = 0
SR[SR > 1.] = 1.
psnr = PSNR(SR, HR)
ssim = SSIM(SR, HR)
sam = SAM(SR, HR)
PSNRs.append(psnr)
SSIMs.append(ssim)
SAMs.append(sam)
SR = SR.transpose(1, 2, 0)
HR = HR.transpose(1, 2, 0)
scio.savemat(out_path + images_name[index], {'HR': HR, 'SR': SR})
print(
"===The {}-th picture=====PSNR:{:.3f}=====SSIM:{:.4f}=====SAM:{:.3f}====Name:{}"
.format(index + 1, psnr, ssim, sam, images_name[index]))
print("=====averPSNR:{:.3f}=====averSSIM:{:.4f}=====averSAM:{:.3f}".format(
np.mean(PSNRs), np.mean(SSIMs), np.mean(SAMs)))
print T / len(images_name)
def val(val_loader, model, epoch):
model.eval()
val_psnr = 0
val_ssim = 0
val_sam = 0
for iteration, batch in enumerate(val_loader, 1):
input, bicu, HR = Variable(batch[0], volatile=True), Variable(
batch[1], volatile=True), Variable(batch[2])
SR = np.zeros(
(HR.shape[1], HR.shape[2], HR.shape[3])).astype(np.float32)
first_state = None
if opt.cuda:
input = input.cuda()
bicu = bicu.cuda()
neigbor_l = []
neigbor_r = []
for i in range(input.shape[1]):
if i == 0:
neigbor_l.append(input[:, 3, :, :].data.unsqueeze(1))
neigbor_l.append(input[:, 1, :, :].data.unsqueeze(1))
neigbor_r.append(input[:, 2, :, :].data.unsqueeze(1))
neigbor_r.append(input[:, 4, :, :].data.unsqueeze(1))
elif i == 1:
neigbor_l.append(input[:, 3, :, :].data.unsqueeze(1))
neigbor_l.append(input[:, 0, :, :].data.unsqueeze(1))
neigbor_r.append(input[:, 2, :, :].data.unsqueeze(1))
neigbor_r.append(input[:, 4, :, :].data.unsqueeze(1))
elif i == input.shape[1] - 2:
neigbor_l.append(input[:, i - 3, :, :].data.unsqueeze(1))
neigbor_l.append(input[:, i - 1, :, :].data.unsqueeze(1))
neigbor_r.append(input[:, i + 1, :, :].data.unsqueeze(1))
neigbor_r.append(input[:, i - 2, :, :].data.unsqueeze(1))
elif i == input.shape[1] - 1:
neigbor_l.append(input[:, i - 3, :, :].data.unsqueeze(1))
neigbor_l.append(input[:, i - 1, :, :].data.unsqueeze(1))
neigbor_r.append(input[:, i - 2, :, :].data.unsqueeze(1))
neigbor_r.append(input[:, i - 4, :, :].data.unsqueeze(1))
else:
neigbor_l.append(input[:, i - 2, :, :].data.unsqueeze(1))
neigbor_l.append(input[:, i - 1, :, :].data.unsqueeze(1))
neigbor_r.append(input[:, i + 1, :, :].data.unsqueeze(1))
neigbor_r.append(input[:, i + 2, :, :].data.unsqueeze(1))
x = input[:, i, :, :]
single_bicu = bicu[:, i, :, :]
left = Variable(torch.cat(neigbor_l, 1))
right = Variable(torch.cat(neigbor_r, 1))
output, first_state = model(i, x, single_bicu, left, right,
first_state)
SR[i, :, :] = output.cpu().data[0].numpy()
val_psnr += PSNR(SR, HR.data[0].numpy())
val_ssim += SSIM(SR, HR.data[0].numpy())
val_sam += SAM(SR, HR.data[0].numpy())
print("PSNR = {:.3f} SSIM = {:.4F} SAM = {:.3f}".format(
val_psnr / len(val_loader), val_ssim / len(val_loader),
val_sam / len(val_loader)))
if opt.show:
writer.add_scalar('Val/PSNR', val_psnr, epoch)
def main():
input_path = '/media/hdisk/liqiang/hyperSR/test/' + opt.datasetName + '/' + str(
opt.upscale_factor) + '/'
out_path = '/media/hdisk/liqiang/hyperSR/result/' + opt.datasetName + '/' + str(
opt.upscale_factor) + '/' + opt.method + '/'
if not os.path.exists(out_path):
os.makedirs(out_path)
PSNRs = []
SSIMs = []
SAMs = []
if not os.path.exists(out_path):
os.makedirs(out_path)
if opt.cuda:
print("=> use gpu id: '{}'".format(opt.gpus))
os.environ["CUDA_VISIBLE_DEVICES"] = opt.gpus
if not torch.cuda.is_available():
raise Exception(
"No GPU found or Wrong gpu id, please run without --cuda")
model = pre_Net(opt)
if opt.cuda:
model = nn.DataParallel(model).cuda()
checkpoint = torch.load(opt.model_name)
model.load_state_dict(checkpoint['model'])
model.eval()
images_name = [x for x in listdir(input_path) if is_image_file(x)]
for index in range(len(images_name)):
mat = scio.loadmat(input_path + images_name[index])
LR = mat['LR'].astype(np.float32).transpose(2, 0, 1)
HR = mat['HR'].astype(np.float32).transpose(2, 0, 1)
bic = np.zeros(HR.shape, dtype=np.float32)
for i in range(bic.shape[0]):
bic[i, :, :] = imresize(LR[i, :, :], (bic.shape[1], bic.shape[2]),
'bicubic',
mode='F')
bicu = Variable(torch.from_numpy(bic).float(),
volatile=True).contiguous().view(
1, -1, bic.shape[1], bic.shape[2])
input = Variable(torch.from_numpy(LR).float(),
volatile=True).contiguous().view(
1, -1, LR.shape[1], LR.shape[2])
SR = np.array(HR).astype(np.float32)
first_state = None
if opt.cuda:
input = input.cuda()
bicu = bicu.cuda()
neigbor_l = []
neigbor_r = []
for i in range(input.shape[1]):
if i == 0:
neigbor_l.append(input[:, 3, :, :].data.unsqueeze(1))
neigbor_l.append(input[:, 1, :, :].data.unsqueeze(1))
neigbor_r.append(input[:, 2, :, :].data.unsqueeze(1))
neigbor_r.append(input[:, 4, :, :].data.unsqueeze(1))
elif i == 1:
neigbor_l.append(input[:, 3, :, :].data.unsqueeze(1))
neigbor_l.append(input[:, 0, :, :].data.unsqueeze(1))
neigbor_r.append(input[:, 2, :, :].data.unsqueeze(1))
neigbor_r.append(input[:, 4, :, :].data.unsqueeze(1))
elif i == input.shape[1] - 2:
neigbor_l.append(input[:, i - 3, :, :].data.unsqueeze(1))
neigbor_l.append(input[:, i - 1, :, :].data.unsqueeze(1))
neigbor_r.append(input[:, i + 1, :, :].data.unsqueeze(1))
neigbor_r.append(input[:, i - 2, :, :].data.unsqueeze(1))
elif i == input.shape[1] - 1:
neigbor_l.append(input[:, i - 3, :, :].data.unsqueeze(1))
neigbor_l.append(input[:, i - 1, :, :].data.unsqueeze(1))
neigbor_r.append(input[:, i - 2, :, :].data.unsqueeze(1))
neigbor_r.append(input[:, i - 4, :, :].data.unsqueeze(1))
else:
neigbor_l.append(input[:, i - 2, :, :].data.unsqueeze(1))
neigbor_l.append(input[:, i - 1, :, :].data.unsqueeze(1))
neigbor_r.append(input[:, i + 1, :, :].data.unsqueeze(1))
neigbor_r.append(input[:, i + 2, :, :].data.unsqueeze(1))
x = input[:, i, :, :]
single_bicu = bicu[:, i, :, :]
left = Variable(torch.cat(neigbor_l, 1))
right = Variable(torch.cat(neigbor_r, 1))
output, first_state = model(i, x, single_bicu, left, right,
first_state)
SR[i, :, :] = output.cpu().data[0].numpy()
SR[SR < 0] = 0
SR[SR > 1.] = 1.
psnr = PSNR(SR, HR)
ssim = SSIM(SR, HR)
sam = SAM(SR, HR)
PSNRs.append(psnr)
SSIMs.append(ssim)
SAMs.append(sam)
SR = SR.transpose(1, 2, 0)
HR = HR.transpose(1, 2, 0)
scio.savemat(out_path + images_name[index], {'HR': HR, 'SR': SR})
print(
"===The {}-th picture=====PSNR:{:.3f}=====SSIM:{:.4f}=====SAM:{:.3f}====Name:{}"
.format(index + 1, psnr, ssim, sam, images_name[index]))
print("=====averPSNR:{:.3f}=====averSSIM:{:.4f}=====averSAM:{:.3f}".format(
np.mean(PSNRs), np.mean(SSIMs), np.mean(SAMs)))
def main():
input_path = '/mnt/liguanlin/DataSets/hypserdatasets/CAVE/test/2/'
out_path = '/mnt/liguanlin/DataSets/hypserdatasets/CAVE/result/2/'
PSNRs = []
SSIMs = []
SAMs = []
if not os.path.exists(out_path):
os.makedirs(out_path)
if opt.cuda:
print("=> use gpu id: '{}'".format(opt.gpus))
os.environ["CUDA_VISIBLE_DEVICES"] = opt.gpus
if not torch.cuda.is_available():
raise Exception(
"No GPU found or Wrong gpu id, please run without --cuda")
model = MCNet(opt)
model.eval()
if opt.cuda:
model = nn.DataParallel(model).cuda()
checkpoint = torch.load(opt.model_name)
model.load_state_dict(checkpoint["model"])
images_name = [x for x in listdir(input_path) if is_image_file(x)]
for index in range(len(images_name)):
mat = scio.loadmat(input_path + images_name[index])
hyperLR = mat['LR'].transpose(2, 0, 1).astype(np.float32)
input = Variable(torch.from_numpy(hyperLR).float(),
volatile=True).contiguous().view(
1, -1, hyperLR.shape[1], hyperLR.shape[2])
if opt.cuda:
input = input.cuda()
with torch.no_grad(): #加上这一句，否则会出现gpu内存不足的提示
output = model(input)
HR = mat['HR'].transpose(2, 0, 1).astype(np.float32)
SR = output.cpu().data[0].numpy().astype(np.float32)
SR[SR < 0] = 0
SR[SR > 1.] = 1.
print('SR.shape:', SR.shape)
print('HR.shape:', HR.shape)
psnr = PSNR(SR, HR)
ssim = SSIM(SR, HR)
sam = SAM(SR, HR)
PSNRs.append(psnr)
SSIMs.append(ssim)
SAMs.append(sam)
SR = SR.transpose(1, 2, 0)
HR = HR.transpose(1, 2, 0)
scio.savemat(out_path + images_name[index], {'HR': HR, 'SR': SR})
print(
"===The {}-th picture=====PSNR:{:.3f}=====SSIM:{:.4f}=====SAM:{:.3f}====Name:{}"
.format(
index + 1,
psnr,
ssim,
sam,
images_name[index],
))
print("=====averPSNR:{:.3f}=====averSSIM:{:.4f}=====averSAM:{:.3f}".format(
np.mean(PSNRs), np.mean(SSIMs), np.mean(SAMs)))