def __getitem__(self, index):
filename_hr = os.path.join(self.data_root, self.datasets_hr.iloc[index,
0])
# # PIL read
image_hr = Image.open(filename_hr).convert('RGB')
image_hr = self.transform(image_hr)
image_hr = image_hr.permute(1, 2, 0)
image_hr = image_hr * 255
# cv2 read
# image_hr = cv2.imread(filename_hr)
# image_hr = torch.from_numpy(image_hr)
# 0-255 cv2(256,256,3) (RGB)
# image_hr, image_lr, filename_lr = lr_data(self.imgfilter_l, self.imgfilter_h, self.with_possion, self.with_texturesyn_thesis,image_hr, filename_hr, self.noises_mean, self.noise_yuv, 1)
image_hr, image_lr, filename_lr = lr_data(
self.imgfilter_l, self.imgfilter_h, self.with_possion,
self.with_texturesyn_thesis, image_hr, filename_hr,
self.noises_mean, self.noise_yuv, 1)
image_hr = torch.clamp(image_hr / 255, 0, 1)
image_lr = torch.clamp(image_lr / 255, 0, 1)
data = {'lr': image_lr, 'hr': image_hr, 'filename': filename_lr}
# RGB tensor (0,1)
# # test
# hhr = image_hr.numpy()
# hhr = cv2.cvtColor(hhr, cv2.COLOR_RGB2BGR)
# cv2.imwrite(self.main_path + "hr" + ".jpg", hhr*255 )
# llr = image_lr.numpy()
# llr = cv2.cvtColor(llr, cv2.COLOR_RGB2BGR)
# cv2.imwrite(self.main_path + "lr" + ".jpg", llr*255 )
return data
def __getitem__(self, index):
filename_hr = os.path.join(self.data_root, self.datasets_hr.iloc[index,
0])
image_hr = cv2.imread(filename_hr)
image_hr = torch.from_numpy(image_hr)
image_hr, image_lr, filename_lr = lr_data(image_hr, filename_hr,
self.noises_mean,
self.noise_yuv, 1)
## test
# cv2.imwrite("./test/" + "hr" + ".jpg", image_hr)
# cv2.imwrite("./test/" + "lr" + ".jpg", image_lr)
# image_lr = image_lr.to(self.device)
# image_hr = image_hr.to(self.device)
data = {'lr': image_lr, 'hr': image_hr, 'filename': filename_lr}
# data = [image_lr, image_hr]
return data
def __getitem__(self, index):
filename_hr = os.path.join(self.data_root, self.datasets_hr.iloc[index,
0])
# # PIL read
image_hr = Image.open(filename_hr).convert('RGB')
image_hr = self.transform(image_hr)
image_hr = image_hr.permute(1, 2, 0)
image_hr = image_hr * 255
# cv2 read
# image_hr = cv2.imread(filename_hr)
# image_hr = torch.from_numpy(image_hr)
image_hr, image_lr, filename_lr = lr_data(
self.imgfilter_l, self.imgfilter_h, self.with_possion,
self.with_texturesyn_thesis, image_hr, filename_hr,
self.noises_mean, self.noise_yuv, 1)
image_hr = torch.clamp(image_hr / 255, 0, 1) #high resolution
image_lr = torch.clamp(image_lr / 255, 0, 1) #low resolution
## add gauss blur
hf = image_hr - torch.from_numpy(
cv2.GaussianBlur(image_hr.numpy(),
(self.gauss_kernel, self.gauss_kernel), 0))
data = {
'lr': image_lr,
'hr': image_hr,
'hf': hf,
'filename': filename_lr
}
# # test
# hhr = image_hr.numpy()
# hhr = cv2.cvtColor(hhr, cv2.COLOR_RGB2BGR)
# cv2.imwrite("./hr" + ".jpg", hhr*255 )
# llr = image_lr.numpy()
# llr = cv2.cvtColor(llr, cv2.COLOR_RGB2BGR)
# cv2.imwrite( "./lr" + ".jpg", llr*255 )
# hhf = hf.numpy()
# hhf = cv2.cvtColor(hhf, cv2.COLOR_RGB2BGR)
# cv2.imwrite("./hf_5" + ".jpg", hhf * 255*10)
return data
def __getitem__(self, index):
filename_hr = os.path.join(self.data_root, self.datasets_hr.iloc[index,
0])
# # PIL read
image_hr = Image.open(filename_hr).convert('RGB')
image_hr = self.transform(image_hr)
# noise_yuv = self.transform(self.noise_yuv)
# noise_yuv = [self.transform(n) for n in self.noise_yuv]
image_hr = image_hr.permute(1, 2, 0)
image_hr = image_hr * 255
# noise_yuv = noise_yuv*255
# cv2 read
# image_hr = cv2.imread(filename_hr)
# image_hr = torch.from_numpy(image_hr)
image_hr, image_lr, filename_lr = lr_data(image_hr, filename_hr,
self.noises_mean,
self.noise_yuv, 1)
## test
# cv2.imwrite(self.data_root + "hr" + ".jpg", image_hr)
# cv2.imwrite("./test/" + "lr" + ".jpg", image_lr)
# image_hr = image_hr.numpy()
# image_lr = image_lr.numpy()
# image_hr = Image.fromarray(cv2.cvtColor(image_hr, cv2.COLOR_BGR2RGB).astype(np.uint8))
# image_lr = Image.fromarray(cv2.cvtColor(image_lr, cv2.COLOR_BGR2RGB).astype(np.uint8))
# #
# image_hr = self.transform(image_hr)
# image_lr = self.transform(image_lr)
# image_hr = image_hr.permute(1, 2, 0)
# image_lr = image_lr.permute(1, 2, 0)
# image_hr = torch.from_numpy(image_hr)
# image_lr = torch.from_numpy(image_lr)
data = {'lr': image_lr, 'hr': image_hr, 'filename': filename_lr}
return data