fix #4

CLOTHING1M/load_cloth1m_data.py

@@ -0,0 +1,62 @@
+from PIL import Image
+import os
+import os.path
+import numpy as np
+import torch
+import torch.utils.data as data
+import torchvision.transforms as transforms
+
+def create_img_folder(img_list, label_list, folder_name, root='../data/clothing1M'):
+    # load label dict
+    label_dict = {}
+    with open(os.path.join(root, label_list), 'r') as f:
+        for line in f:
+            parts = line.strip().split(' ')
+            label_dict[parts[0]] = int(parts[1])
+
+    # following previous works on cloth1m
+    preprocess_transform = transforms.Compose([
+        transforms.Resize((256,256)),
+        transforms.CenterCrop(224),
+    ])
+
+    img_tensor_list = []
+    label_list = []
+    file_list = []
+
+    # create folder if not exist
+    if not os.path.isdir(os.path.join(root, folder_name)):
+        os.mkdir(os.path.join(root, folder_name))
+    
+    cnt = 0
+    name_set = set()
+    with open(os.path.join(root, img_list), 'r') as f:
+        for line in f:
+            img_name = line.strip()
+            label = label_dict[img_name]
+
+            out_dir = os.path.join(root, folder_name, str(label))
+            if not os.path.isdir(out_dir):
+                os.mkdir(out_dir)
+
+            trailing_name = img_name.split('/')[-1]
+            assert trailing_name not in name_set, 'Image duplicates!'
+            name_set.add(trailing_name)
+            
+            with Image.open(os.path.join(root, img_name)) as img:
+                processed_img = preprocess_transform(img)
+
+            processed_img.save(os.path.join(out_dir, trailing_name))
+            cnt += 1
+
+            if (cnt % 10000 == 0):
+                print ('%d images processed' % cnt)
+
+    print ('In total: %d images processed.' % cnt)
+
+
+if __name__ == '__main__':
+    create_img_folder('noisy_train_key_list.txt', 'noisy_label_kv.txt', 'noisy_train', root='../data/clothing1M')
+    create_img_folder('clean_train_key_list.txt', 'clean_label_kv.txt', 'clean_train', root='../data/clothing1M')
+    create_img_folder('clean_val_key_list.txt',   'clean_label_kv.txt', 'clean_val', root='../data/clothing1M')
+    create_img_folder('clean_test_key_list.txt',  'clean_label_kv.txt', 'clean_test', root='../data/clothing1M')