IAT에 데이터셋 이미지 크기 일괄적으로 변경하기

Image.open()으로 하니까 PIL resize라고 검색ㅎㅐ서 resize하는 방법을 data_lowlight = data_lowlight.resize(self.resize), data_highlight = data_highlight.resize(self.resize) 을 if self.mode == 'train': 과 elif self.mode == 'test': 에 각각 추가

    def __getitem__(self, index):
        data_lowlight_path = self.data_list[index]
        
        if self.mode == 'train':
            data_lowlight = Image.open(data_lowlight_path)
            data_lowlight = data_lowlight.resize(self.resize)
            data_highlight = Image.open(data_lowlight_path.replace('low', 'normal').replace('Low','Normal'))
            data_highlight = data_highlight.resize(self.resize)
            
            data_lowlight, data_highlight = self.FLIP_LR(data_lowlight, data_highlight)
            data_lowlight, data_highlight = self.FLIP_UD(data_lowlight, data_highlight)
            data_lowlight, data_highlight = self.Random_Crop(data_lowlight, data_highlight)
            
            # print(self.w, self.h)
            #print(data_lowlight.size, data_highlight.size)
            
            data_lowlight = data_lowlight.resize((self.w, self.h), Image.ANTIALIAS)
            data_highlight = data_highlight.resize((self.w, self.h), Image.ANTIALIAS)
            data_lowlight, data_highlight = (np.asarray(data_lowlight) / 255.0), (np.asarray(data_highlight) / 255.0)

            if self.normalize:
                #data_lowlight, data_highlight = torch.from_numpy(data_lowlight).float(), torch.from_numpy(data_highlight).float()
                transform_input = Compose([ToTensor(), Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5)), ConvertImageDtype(torch.float), ])
                transform_gt = Compose([ToTensor(), ConvertImageDtype(torch.float), ])
                #return transform_input(data_lowlight).permute(2, 0, 1), transform_gt(data_highlight).permute(2, 0, 1)
                return transform_input(data_lowlight), transform_gt(data_highlight)
            else:
                data_lowlight, data_highlight = torch.from_numpy(data_lowlight).float(), torch.from_numpy(data_highlight).float()
                return data_lowlight.permute(2,0,1), data_highlight.permute(2,0,1)

        elif self.mode == 'test':
            data_lowlight = Image.open(data_lowlight_path)
            data_lowlight = data_lowlight.resize(self.resize)
            data_highlight = Image.open(data_lowlight_path.replace('low', 'normal').replace('Low','Normal'))
            data_highlight = data_highlight.resize(self.resize)
            data_lowlight, data_highlight = (np.asarray(data_lowlight) / 255.0), (np.asarray(data_highlight) / 255.0)
            #data_lowlight, data_highlight = torch.from_numpy(data_lowlight).float(), torch.from_numpy(data_highlight).float()
            if self.normalize:
                #data_lowlight, data_highlight = torch.from_numpy(data_lowlight).float(), torch.from_numpy(data_highlight).float()
                transform_input = Compose([ToTensor(), Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5)), ConvertImageDtype(torch.float), ])
                transform_gt = Compose([ToTensor(), ConvertImageDtype(torch.float), ])
                #return transform_input(data_lowlight).permute(2, 0, 1), transform_gt(data_highlight).permute(2, 0, 1)
                return transform_input(data_lowlight), transform_gt(data_highlight)
            else:
                data_lowlight, data_highlight = torch.from_numpy(data_lowlight).float(), torch.from_numpy(data_highlight).float()
                return data_lowlight.permute(2,0,1), data_highlight.permute(2,0,1)

data_lowlight.resize(self.resize)

class lowlight_loader(data.Dataset):

    def __init__(self, images_path, mode='train', normalize=True, resize=(512,512)):
        self.train_list = populate_train_list(images_path, mode)
        #self.h, self.w = int(img_size[0]), int(img_size[1])
        # train or test
        self.mode = mode
        self.data_list = self.train_list
        self.normalize = normalize
        self.resize = resize
        print("Total examples:", len(self.train_list))

self.resize = resize 추가함, resize=(512,512) 디폴트 값도 추가.