feat: 코드 업로드 및 최종보고서, ppt 업로드

@@ -45,9 +45,8 @@ from pycocotools.mask import encode
 import argparse
 import argparse
 
 
 parser = argparse.ArgumentParser(description="PyTorch CARN")
 parser = argparse.ArgumentParser(description="PyTorch CARN")
-parser.add_argument("--data_path", type=str, default = "/home/ubuntu/JH/exp1/dataset")
 parser.add_argument("--valid_data_path", type=str)
 parser.add_argument("--valid_data_path", type=str)
-parser.add_argument("--rescale_factor", type=int, default=4, help="rescale factor for using in training")
+parser.add_argument("--rescale_factor", type=int, help="rescale factor for using in training")
 parser.add_argument("--model_name", type=str,choices= ["VDSR", "CARN", "SRRN","FRGAN"], default='CARN', help="Feature type for usingin training")
 parser.add_argument("--model_name", type=str,choices= ["VDSR", "CARN", "SRRN","FRGAN"], default='CARN', help="Feature type for usingin training")
 parser.add_argument("--loss_type", type=str, choices= ["MSE", "L1", "SmoothL1","vgg_loss","ssim_loss","adv_loss","lpips"], default='MSE', help="loss type in training")
 parser.add_argument("--loss_type", type=str, choices= ["MSE", "L1", "SmoothL1","vgg_loss","ssim_loss","adv_loss","lpips"], default='MSE', help="loss type in training")
 parser.add_argument('--batch_size', type=int, default=256)
 parser.add_argument('--batch_size', type=int, default=256)
@@ -153,9 +152,12 @@ for iter in range(0, 100):
 
 
   globals()['maxRange_{}'.format(image_file_number)] = maxRange
   globals()['maxRange_{}'.format(image_file_number)] = maxRange
 
 
-  # p2_feature_img = Image.open('./original/qp32/COCO_val2014_000000'+ image_file_number +'_p2.png'
+  p2_feature_img = Image.open('/content/drive/MyDrive/result/inference_x{}/LR_2/p2/'.format(opt.rescale_factor) + image_prefix + '000000' + image_file_number + '_p2' +'.png')
+  p3_feature_img = Image.open('/content/drive/MyDrive/result/inference_x{}/LR_2/p3/'.format(opt.rescale_factor) + image_prefix + '000000' + image_file_number + '_p3'  +'.png')
+  p4_feature_img = Image.open('/content/drive/MyDrive/result/inference_x{}/LR_2/p4/'.format(opt.rescale_factor) + image_prefix + '000000' + image_file_number + '_p4' +'.png')
+# p2_feature_img = Image.open('./original/qp32/COCO_val2014_000000'+ image_file_number +'_p2.png'
   # p2_feature_img = Image.open('./result/{}/inference/{}_p2x{}/SR_{}.png'.format(opt.loss_type,opt.model_name,opt.rescale_factor,str(iter)))
   # p2_feature_img = Image.open('./result/{}/inference/{}_p2x{}/SR_{}.png'.format(opt.loss_type,opt.model_name,opt.rescale_factor,str(iter)))
-  p2_feature_img = Image.open('/content/drive/MyDrive/result/inference/LR_2/p2/' + image_prefix + '000000' + image_file_number + '_p2' +'.png')
+  
   # # y_p2, cb, cr = p2_feature_img.split()
   # # y_p2, cb, cr = p2_feature_img.split()
   p2_feature_arr = np.array(p2_feature_img)
   p2_feature_arr = np.array(p2_feature_img)
   p2_feature_arr_round = myRound(p2_feature_arr)
   p2_feature_arr_round = myRound(p2_feature_arr)
@@ -163,14 +165,14 @@ for iter in range(0, 100):
   # p3_feature_img = Image.open('./original/qp32/COCO_val2014_000000'+ image_file_number +'_p3.png')
   # p3_feature_img = Image.open('./original/qp32/COCO_val2014_000000'+ image_file_number +'_p3.png')
 
 
   # p3_feature_img = Image.open('./result/{}/inference/{}_p3x{}/SR_{}.png'.format(opt.loss_type,opt.model_name,opt.rescale_factor,str(iter)))
   # p3_feature_img = Image.open('./result/{}/inference/{}_p3x{}/SR_{}.png'.format(opt.loss_type,opt.model_name,opt.rescale_factor,str(iter)))
-  p3_feature_img = Image.open('/content/drive/MyDrive/result/inference/LR_2/p3/' + image_prefix + '000000' + image_file_number + '_p3'  +'.png')
+  
   # # y_p3, cb2, cr2 = p3_feature_img.split()
   # # y_p3, cb2, cr2 = p3_feature_img.split()
   p3_feature_arr = np.array(p3_feature_img)
   p3_feature_arr = np.array(p3_feature_img)
   p3_feature_arr_round = myRound(p3_feature_arr)
   p3_feature_arr_round = myRound(p3_feature_arr)
 
 
   # p4_feature_img = Image.open('./original/qp32/COCO_val2014_000000'+ image_file_number +'_p4.png')
   # p4_feature_img = Image.open('./original/qp32/COCO_val2014_000000'+ image_file_number +'_p4.png')
   # p4_feature_img = Image.open('./result/{}/inference/{}_p4x{}/SR_{}.png'.format(opt.loss_type,opt.model_name,opt.rescale_factor,str(iter)))
   # p4_feature_img = Image.open('./result/{}/inference/{}_p4x{}/SR_{}.png'.format(opt.loss_type,opt.model_name,opt.rescale_factor,str(iter)))
-  p4_feature_img = Image.open('/content/drive/MyDrive/result/inference/LR_2/p4/' + image_prefix + '000000' + image_file_number + '_p4' +'.png')
+  
   # y_p4, cb3, cr3 = p4_feature_img.split()
   # y_p4, cb3, cr3 = p4_feature_img.split()
   p4_feature_arr = np.array(p4_feature_img)
   p4_feature_arr = np.array(p4_feature_img)
   p4_feature_arr_round = myRound(p4_feature_arr)
   p4_feature_arr_round = myRound(p4_feature_arr)

@@ -117,9 +117,7 @@ model = torch.load(opt.model, map_location=lambda storage, loc: storage)["model"
 
 
 scales = [opt.scaleFactor]
 scales = [opt.scaleFactor]
 
 
-# image_list = glob.glob(opt.dataset+"/*.*") 
 if opt.singleImage == "Y" :
 if opt.singleImage == "Y" :
-    # image_list = crop_feature(opt.dataset, opt.featureType, opt.scaleFactor)
     image_list = opt.dataset
     image_list = opt.dataset
 else:
 else:
     image_path = os.path.join(opt.dataset, opt.featureType)
     image_path = os.path.join(opt.dataset, opt.featureType)
@@ -150,7 +148,6 @@ for scale in scales:
             f_bi = f_bi.astype(float)
             f_bi = f_bi.astype(float)
             features_bicubic.append(f_bi)
             features_bicubic.append(f_bi)
             psnr_bicubic = PSNR(f_bi, f_gt, shave_border=scale)
             psnr_bicubic = PSNR(f_bi, f_gt, shave_border=scale)
-            # psnr_bicubic = PSNR_ver2(cv2.imread(f_gt), cv2.imread(f_bi))
             avg_psnr_bicubic += psnr_bicubic
             avg_psnr_bicubic += psnr_bicubic
 
 
             f_input = f_bi/255.
             f_input = f_bi/255.
@@ -177,7 +174,6 @@ for scale in scales:
             f_sr = f_sr[0,:,:]
             f_sr = f_sr[0,:,:]
 
 
             psnr_predicted = PSNR(f_sr, f_gt, shave_border=scale)
             psnr_predicted = PSNR(f_sr, f_gt, shave_border=scale)
-            # psnr_predicted = PSNR_ver2(cv2.imread(f_gt), cv2.imread(f_sr))
             avg_psnr_predicted += psnr_predicted
             avg_psnr_predicted += psnr_predicted
             features.append(f_sr)
             features.append(f_sr)
 
 
@@ -187,23 +183,3 @@ for scale in scales:
     print("Dataset=", opt.dataset)
     print("Dataset=", opt.dataset)
     print("Average PSNR_predicted=", avg_psnr_predicted/count)
     print("Average PSNR_predicted=", avg_psnr_predicted/count)
     print("Average PSNR_bicubic=", avg_psnr_bicubic/count)
     print("Average PSNR_bicubic=", avg_psnr_bicubic/count)
-
-
-# Show graph
-# f_gt = Image.fromarray(f_gt)
-# f_b = Image.fromarray(f_bi)
-# f_sr = Image.fromarray(f_sr)
-
-# fig = plt.figure(figsize=(18, 16), dpi= 80)
-# ax = plt.subplot("131")
-# ax.imshow(f_gt)
-# ax.set_title("GT")
-
-# ax = plt.subplot("132")
-# ax.imshow(f_bi)
-# ax.set_title("Input(bicubic)")
-
-# ax = plt.subplot("133")
-# ax.imshow(f_sr)
-# ax.set_title("Output(vdsr)")
-# plt.show()

@@ -13,7 +13,7 @@ from datasets import get_training_data_loader
 # from feature_dataset import get_training_data_loader
 # from feature_dataset import get_training_data_loader
 # from make_dataset import make_dataset
 # from make_dataset import make_dataset
 import numpy as np
 import numpy as np
-from dataFromH5 import Read_dataset_h5
+# from dataFromH5 import Read_dataset_h5
 import matplotlib.pyplot as plt
 import matplotlib.pyplot as plt
 import math
 import math
 
 

+import torch.utils.data as data
+import torch
+import h5py
+
+class Read_dataset_h5(data.Dataset):
+    def __init__(self, file_path):
+        super(Read_dataset_h5, self).__init__()
+        hf = h5py.File(file_path)
+        self.input = hf.get('input')
+        self.label = hf.get('label')
+
+    def __getitem__(self, index):
+        return torch.from_numpy(self.input[index,:,:,:]).float(), torch.from_numpy(self.label[index,:,:,:]).float()
+
+    def __len__(self):
+        return self.input.shape[0]

+'''
+현재 사용하지 않음.
+'''
+
+# from torch.utils.data import Dataset
+# from PIL import Image
+# import os
+# from glob import glob
+# from torchvision import transforms
+# from torch.utils.data.dataset import Dataset
+# import torch
+# import pdb
+# import math
+# import numpy as np
+# import cv2
+
+
+# class FeatureDataset(Dataset):
+#     def __init__(self, data_path, datatype, rescale_factor, valid):
+#         self.data_path = data_path
+#         self.datatype = datatype
+#         self.rescale_factor = rescale_factor
+#         if not os.path.exists(data_path):
+#             raise Exception(f"[!] {self.data_path} not existed")
+#         if (valid):
+#             self.hr_path = os.path.join(self.data_path, 'valid')
+#             self.hr_path = os.path.join(self.hr_path, self.datatype)
+#         else:
+#             self.hr_path = os.path.join(self.data_path, 'LR_2')
+#             self.hr_path = os.path.join(self.hr_path, self.datatype)
+#         print(self.hr_path)
+#         self.names = os.listdir(self.hr_path)
+#         self.hr_path = sorted(glob(os.path.join(self.hr_path, "*.*")))
+#         self.hr_imgs = []
+        
+#         w, h = Image.open(self.hr_path[0]).size
+#         self.width = int(w / 16)
+#         self.height = int(h / 16)
+#         self.lwidth = int(self.width / self.rescale_factor) # rescale_factor만큼 크기를 줄인다.
+#         self.lheight = int(self.height / self.rescale_factor)
+#         print("lr: ({} {}), hr: ({} {})".format(self.lwidth, self.lheight, self.width, self.height))
+
+#         self.original_hr_imgs = [] #원본 250개
+#         print("crop features ...")
+#         for hr in self.hr_path: # 256개의 피쳐로 나눈다.
+#             hr_cropped_imgs = []
+#             hr_image = Image.open(hr)  # .convert('RGB')\
+#             self.original_hr_imgs.append(np.array(hr_image).astype(float)) # 원본을 저장한다.
+#             for i in range(16):
+#                 for j in range(16):
+#                     (left, upper, right, lower) = (
+#                     i * self.width, j * self.height, (i + 1) * self.width, (j + 1) * self.height)
+#                     crop = hr_image.crop((left, upper, right, lower))
+#                     hr_cropped_imgs.append(crop)
+#             self.hr_imgs.append(hr_cropped_imgs)
+
+#         self.final_results = [] # [250개]
+#         print("resize and concat features ...")
+#         for i in range(0, len(self.hr_imgs)):
+#             hr_img = self.hr_imgs[i]  
+#             interpolated_images = []
+#             for img in hr_img:
+#                 image = img.resize((self.lwidth, self.lheight), Image.BICUBIC)
+#                 image = image.resize((self.width, self.height), Image.BICUBIC)
+#                 interpolated_images.append(np.array(image).astype(float))
+#             self.final_results.append(concatFeatures(interpolated_images, self.names[i], self.datatype))
+#         print(self.original_hr_imgs)
+#         print(self.final_results)
+
+#     def __getitem__(self, idx):
+#       ground_truth = self.original_hr_imgs[idx]
+#       final_result = self.final_results[idx] # list
+#       return transforms.ToTensor()(final_result), transforms.ToTensor()(ground_truth) # hr_image를 변환한 것과, 변환하지 않은 것을 Tensor로 각각 반환
+
+#     def __len__(self):
+#         return len(self.hr_path)
+
+
+# def concatFeatures(features, image_name, feature_type):
+#     features_0 = features[:16]
+#     features_1 = features[16:32]
+#     features_2 = features[32:48]
+#     features_3 = features[48:64]
+#     features_4 = features[64:80]
+#     features_5 = features[80:96]
+#     features_6 = features[96:112]
+#     features_7 = features[112:128]
+#     features_8 = features[128:144]
+#     features_9 = features[144:160]
+#     features_10 = features[160:176]
+#     features_11 = features[176:192]
+#     features_12 = features[192:208]
+#     features_13 = features[208:224]
+#     features_14 = features[224:240]
+#     features_15 = features[240:256]
+    
+#     features_new = list()
+#     features_new.extend([
+#         concat_vertical(features_0),
+#         concat_vertical(features_1),
+#         concat_vertical(features_2),
+#         concat_vertical(features_3),
+#         concat_vertical(features_4),
+#         concat_vertical(features_5),
+#         concat_vertical(features_6),
+#         concat_vertical(features_7),
+#         concat_vertical(features_8),
+#         concat_vertical(features_9),
+#         concat_vertical(features_10),
+#         concat_vertical(features_11),
+#         concat_vertical(features_12),
+#         concat_vertical(features_13),
+#         concat_vertical(features_14),
+#         concat_vertical(features_15)
+#     ])
+
+#     final_concat_feature = concat_horizontal(features_new)
+
+#     save_path = "features/LR_2/" + feature_type + "/" + image_name
+#     if not os.path.exists("features/"):
+#         os.makedirs("features/")
+#     if not os.path.exists("features/LR_2/"):
+#         os.makedirs("features/LR_2/")
+#     if not os.path.exists("features/LR_2/" + feature_type):
+#         os.makedirs("features/LR_2/" + feature_type)
+#     cv2.imwrite(save_path, final_concat_feature)
+    
+#     return np.array(final_concat_feature).astype(float)
+
+# def concat_horizontal(feature):
+#     result = cv2.hconcat([feature[0], feature[1]])
+#     for i in range(2, len(feature)):
+#         result = cv2.hconcat([result, feature[i]])
+#     return result
+       
+# def concat_vertical(feature):
+#     result = cv2.vconcat([feature[0], feature[1]])
+#     for i in range(2, len(feature)):
+#         result = cv2.vconcat([result, feature[i]])
+#     return result
+
+
+# def get_data_loader_test_version(data_path, feature_type, rescale_factor, batch_size, num_workers):
+#     full_dataset = FeatureDataset(data_path, feature_type, rescale_factor, False)
+#     print("dataset의 사이즈는 {}".format(len(full_dataset)))
+#     for f in full_dataset:
+#         print(type(f))
+
+
+# def get_data_loader(data_path, feature_type, rescale_factor, batch_size, num_workers):
+#     full_dataset = FeatureDataset(data_path, feature_type, rescale_factor, False)
+#     train_size = int(0.9 * len(full_dataset))
+#     test_size = len(full_dataset) - train_size
+#     train_dataset, test_dataset = torch.utils.data.random_split(full_dataset, [train_size, test_size])
+#     torch.manual_seed(3334)
+#     train_loader = torch.utils.data.DataLoader(dataset=train_dataset, batch_size=batch_size, shuffle=True,
+#                                                num_workers=num_workers, pin_memory=False)
+#     test_loader = torch.utils.data.DataLoader(dataset=test_dataset, batch_size=batch_size, shuffle=False,
+#                                               num_workers=num_workers, pin_memory=True)
+
+#     return train_loader, test_loader
+
+
+# def get_training_data_loader(data_path, feature_type, rescale_factor, batch_size, num_workers):
+#     full_dataset = FeatureDataset(data_path, feature_type, rescale_factor, False)
+#     torch.manual_seed(3334)
+#     train_loader = torch.utils.data.DataLoader(dataset=full_dataset, batch_size=batch_size, shuffle=True,
+#                                                num_workers=num_workers, pin_memory=False)
+#     return train_loader
+
+
+# def get_infer_dataloader(data_path, feature_type, rescale_factor, batch_size, num_workers):
+#     dataset = FeatureDataset(data_path, feature_type, rescale_factor, True)
+#     data_loader = torch.utils.data.DataLoader(dataset=dataset, batch_size=batch_size, shuffle=False,
+#                                               num_workers=num_workers, pin_memory=False)
+#     return data_loader
\ No newline at end of file

+'''
+현재 사용하지 않음.
+'''
+
+
+# from crop_feature import crop_feature
+# import os
+# from PIL import Image
+# import numpy as np
+# import torch
+# from torch.utils.data.dataset import Dataset
+# from torch.utils.data import TensorDataset, DataLoader
+# from torchvision import transforms
+# import cv2
+# import glob
+# import h5py
+# import argparse
+
+# parser = argparse.ArgumentParser(description="make Dataset")
+# parser.add_argument("--dataset", type=str)
+# parser.add_argument("--featureType", type=str)
+# parser.add_argument("--scaleFactor", type=int)
+# parser.add_argument("--batchSize", type=int, default=16)
+# parser.add_argument("--threads", type=int, default=3)
+
+# # dataset, feature_type, scale_factor, batch_size, num_workers
+# def main():
+#     opt = parser.parse_args()
+
+#     dataset = opt.dataset
+#     feature_type = opt.featureType
+#     scale_factor = opt.scaleFactor
+#     batch_size = opt.batchSize
+#     num_workers = opt.threads
+
+#     print_message = True
+#     dataset = dataset+"/LR_2"
+#     image_path = os.path.join(dataset, feature_type)
+#     image_list = os.listdir(image_path)
+#     input = list()
+#     label = list()
+
+#     for image in image_list:
+#         origin_image = Image.open(os.path.join(image_path,image))
+#         label.append(np.array(origin_image).astype(float))
+#         image_cropped = crop_feature(os.path.join(image_path, image), feature_type, scale_factor, print_message)
+#         print_message = False
+#         # bicubic interpolation
+#         reconstructed_features = list()
+#         print("crop is done.")
+#         for crop in image_cropped:
+#             w, h = crop.size
+#             bicubic_interpolated_image = crop.resize((w//scale_factor, h//scale_factor), Image.BICUBIC)
+#             bicubic_interpolated_image = bicubic_interpolated_image.resize((w,h), Image.BICUBIC) # 다시 원래 크기로 키우기
+#             reconstructed_features.append(np.array(bicubic_interpolated_image).astype(float))
+#         input.append(concatFeatures(reconstructed_features, image, feature_type))
+
+#     print("concat is done.")
+#     if len(input) == len(label):
+#         save_h5(input, label, 'data/train_{}.h5'.format(feature_type))
+#         print("saved..")
+#     else:
+#         print(len(input), len(label), "이 다릅니다.")
+
+
+# def concatFeatures(features, image_name, feature_type):
+#     features_0 = features[:16]
+#     features_1 = features[16:32]
+#     features_2 = features[32:48]
+#     features_3 = features[48:64]
+#     features_4 = features[64:80]
+#     features_5 = features[80:96]
+#     features_6 = features[96:112]
+#     features_7 = features[112:128]
+#     features_8 = features[128:144]
+#     features_9 = features[144:160]
+#     features_10 = features[160:176]
+#     features_11 = features[176:192]
+#     features_12 = features[192:208]
+#     features_13 = features[208:224]
+#     features_14 = features[224:240]
+#     features_15 = features[240:256]
+    
+#     features_new = list()
+#     features_new.extend([
+#         concat_vertical(features_0),
+#         concat_vertical(features_1),
+#         concat_vertical(features_2),
+#         concat_vertical(features_3),
+#         concat_vertical(features_4),
+#         concat_vertical(features_5),
+#         concat_vertical(features_6),
+#         concat_vertical(features_7),
+#         concat_vertical(features_8),
+#         concat_vertical(features_9),
+#         concat_vertical(features_10),
+#         concat_vertical(features_11),
+#         concat_vertical(features_12),
+#         concat_vertical(features_13),
+#         concat_vertical(features_14),
+#         concat_vertical(features_15)
+#     ])
+
+#     final_concat_feature = concat_horizontal(features_new)
+
+#     save_path = "features/LR_2/" + feature_type + "/" + image_name
+#     if not os.path.exists("features/"):
+#         os.makedirs("features/")
+#     if not os.path.exists("features/LR_2/"):
+#         os.makedirs("features/LR_2/")
+#     if not os.path.exists("features/LR_2/" + feature_type):
+#         os.makedirs("features/LR_2/" + feature_type)
+#     cv2.imwrite(save_path, final_concat_feature)
+    
+#     return np.array(final_concat_feature).astype(float)
+
+# def concat_horizontal(feature):
+#     result = cv2.hconcat([feature[0], feature[1]])
+#     for i in range(2, len(feature)):
+#         result = cv2.hconcat([result, feature[i]])
+#     return result
+       
+# def concat_vertical(feature):
+#     result = cv2.vconcat([feature[0], feature[1]])
+#     for i in range(2, len(feature)):
+#         result = cv2.vconcat([result, feature[i]])
+#     return result
+
+# def save_h5(sub_ip, sub_la, savepath):
+#     if not os.path.exists("data/"):
+#         os.makedirs("data/")    
+    
+#     path = os.path.join(os.getcwd(), savepath)
+#     with h5py.File(path, 'w') as hf:
+#         hf.create_dataset('input', data=sub_ip)
+#         hf.create_dataset('label', data=sub_la)
+
+# if __name__ == "__main__":
+#     main()
\ No newline at end of file