Update dataset/datasets.py

dataset/datasets.py

@@ -1,107 +1,77 @@
-import torch
 import os
+import torch
 import torchvision.transforms as transforms
 import matplotlib.pyplot as plt
+from torch.utils.data import Dataset
+from PIL import Image
 import numpy as np
 
-from utils.utils import generate_mask
-
-
-class TrainDataset(torch.utils.data.Dataset):
-    def __init__(self, data_path, transform = None, mults_amount = 1):
-        self.data = os.listdir(os.path.join(data_path, 'color'))
+class TrainDataset(Dataset):
+    def __init__(self, data_path, transform=None, mults_amount=1):
         self.data_path = data_path
         self.transform = transform
         self.mults_amount = mults_amount
-        
         self.ToTensor = transforms.ToTensor()
+
+        self.file_list = os.listdir(os.path.join(data_path, 'color'))
+
     def __len__(self):
-        return len(self.data)
-    
+        return len(self.file_list)
+
     def __getitem__(self, idx):
-        image_name = self.data[idx]
-        
-        color_img = plt.imread(os.path.join(self.data_path, 'color', image_name))
-        
-
-        if self.mults_amount > 1:
-            mult_number = np.random.choice(range(self.mults_amount))
-            
-            bw_name = image_name[:image_name.rfind('.')] + '_' + str(mult_number) + '.png'
-            dfm_name = image_name[:image_name.rfind('.')] + '_' + str(mult_number) + '_dfm.png'
-        else:
-            bw_name = self.data[idx]
-            dfm_name =  os.path.splitext(self.data[idx])[0] + '0_dfm.png'
-            
-            
-        bw_img =  np.expand_dims(plt.imread(os.path.join(self.data_path, 'bw', bw_name)), 2)
-        dfm_img =  np.expand_dims(plt.imread(os.path.join(self.data_path, 'bw', dfm_name)), 2)
-        
-        bw_img = np.concatenate([bw_img, dfm_img], axis = 2)
-        
+        image_name = self.file_list[idx]
+
+        # Extract mult_number from the image name
+        mult_number = int(image_name.split('_')[-1].split('.')[0])
+
+        # Construct corresponding bw and dfm image names
+        bw_name = image_name.replace(f'_{mult_number}.png', '.png')
+        dfm_name = bw_name.replace('.png', f'_{mult_number}_dfm.png')
+
+        color_img = Image.open(os.path.join(self.data_path, 'color', image_name)).convert('RGB')
+        bw_img = Image.open(os.path.join(self.data_path, 'bw', bw_name)).convert('L')
+        dfm_img = Image.open(os.path.join(self.data_path, 'bw', dfm_name)).convert('L')
+
         if self.transform:
-            result = self.transform(image = color_img, mask = bw_img)
-            color_img = result['image']
-            bw_img = result['mask']
-          
-        dfm_img = bw_img[:, :, 1]
-        bw_img = bw_img[:, :, 0]
-        
-        color_img = self.ToTensor(color_img)
-        bw_img = self.ToTensor(bw_img)
-        
-        dfm_img = self.ToTensor(dfm_img)
-        
-        color_img = (color_img - 0.5) / 0.5
-        
-        mask = generate_mask(bw_img.shape[1], bw_img.shape[2])
-        hint = torch.cat((color_img * mask, mask), 0)
-        
+            color_img = self.transform(color_img)
+            bw_img = self.transform(bw_img)
+            dfm_img = self.transform(dfm_img)
+
+        # Assuming generate_mask is a function generating a mask for hint
+        mask = generate_mask(bw_img.size[1], bw_img.size[0])
+        hint = torch.cat((bw_img * mask, mask), 0)
+
         return bw_img, color_img, hint, dfm_img
-    
-class FineTuningDataset(torch.utils.data.Dataset):
-    def __init__(self, data_path, transform = None, mult_amount = 1):
-        self.data = [x for x in os.listdir(os.path.join(data_path, 'real_manga')) if x.find('_dfm') == -1] 
-        self.color_data = [x for x in os.listdir(os.path.join(data_path, 'color'))]
+
+class FineTuningDataset(Dataset):
+    def __init__(self, data_path, transform=None, mult_amount=1):
         self.data_path = data_path
         self.transform = transform
         self.mults_amount = mult_amount
-        
-        np.random.shuffle(self.color_data)
-        
         self.ToTensor = transforms.ToTensor()
+
+        self.file_list = [x for x in os.listdir(os.path.join(data_path, 'real_manga')) if x.find('_dfm') == -1]
+
     def __len__(self):
-        return len(self.data)
-    
+        return len(self.file_list)
+
     def __getitem__(self, idx):
-        color_img = plt.imread(os.path.join(self.data_path, 'color', self.color_data[idx]))
-        
-        image_name = self.data[idx]
-        if self.mults_amount > 1:
-            mult_number = np.random.choice(range(self.mults_amount))
-            
-            bw_name = image_name[:image_name.rfind('.')] + '_' + str(self.mults_amount) + '.png'
-            dfm_name = image_name[:image_name.rfind('.')] + '_' + str(self.mults_amount) + '_dfm.png'
-        else:
-            bw_name = self.data[idx]
-            dfm_name =  os.path.splitext(self.data[idx])[0] + '_dfm.png'
-        
-        
-        bw_img =  np.expand_dims(plt.imread(os.path.join(self.data_path, 'real_manga', image_name)), 2)
-        dfm_img =  np.expand_dims(plt.imread(os.path.join(self.data_path, 'real_manga', dfm_name)), 2)
-        
+        image_name = self.file_list[idx]
+
+        # Extract mult_number from the image name
+        mult_number = int(image_name.split('_')[-1])
+
+        # Construct corresponding bw and dfm image names
+        bw_name = f"{image_name.split('_')[0]}_{mult_number}.png"
+        dfm_name = f"{image_name.split('_')[0]}_{mult_number}_dfm.png"
+
+        color_img = Image.open(os.path.join(self.data_path, 'color', image_name)).convert('RGB')
+        bw_img = Image.open(os.path.join(self.data_path, 'real_manga', bw_name)).convert('L')
+        dfm_img = Image.open(os.path.join(self.data_path, 'real_manga', dfm_name)).convert('L')
+
         if self.transform:
-            result = self.transform(image = color_img)
-            color_img = result['image']
-            
-            result = self.transform(image = bw_img,  mask = dfm_img)
-            bw_img = result['image']
-            dfm_img = result['mask']
-        
-        color_img = self.ToTensor(color_img)
-        bw_img = self.ToTensor(bw_img)
-        dfm_img = self.ToTensor(dfm_img)
-        
-        color_img = (color_img - 0.5) / 0.5
-        
+            color_img = self.transform(color_img)
+            bw_img = self.transform(bw_img)
+            dfm_img = self.transform(dfm_img)
+
         return bw_img, dfm_img, color_img