| import os |
| from PIL import Image |
| Image.MAX_IMAGE_PIXELS = None |
| import numpy as np |
|
|
| def crop_levir(): |
| ''' |
| Crop LEVIR-CD dataset in original resolution with this folder structure: |
| !!! Make sure the train, test, and val folders are named exactly like what is written below |
| -LEVIR-CD/ |
| ├─train |
| ├─A |
| ├─train_1.png |
| ... |
| ├─B |
| ├─train_1.png |
| ... |
| ├─label |
| ├─train_1.png |
| ├─test |
| ├─A |
| ├─test_1.png |
| ... |
| ├─B |
| ├─test_1.png |
| ... |
| ├─label |
| ├─test_1.png |
| ├─val |
| ├─A |
| ├─val_1.png |
| ... |
| ├─B |
| ├─val_1.png |
| ... |
| ├─label |
| ├─val_1.png |
| ... |
| |
| |
| to 256 x 256 images with this folder structure |
| -LEVIR-Cropped |
| ├─A |
| ├─img1.png |
| ... |
| ├─B |
| ├─img1.png |
| ... |
| ├─label |
| ├─img1.png |
| ... |
| └─list |
| ├─val.txt |
| ├─test.txt |
| └─train.txt |
| ''' |
| |
| ori_folder = r"/mnt/c/Users/prisc/OneDrive - unige.it/Documents/Dataset/LEVIR/LEVIR-CD" |
| |
| cropped_folder = r"/mnt/c/Users/prisc/OneDrive - unige.it/Documents/Dataset/LEVIR/LEVIR-Cropped" |
|
|
| ori_size = 1024 |
| crop_size = 256 |
|
|
| |
| train_list = [] |
| test_list = [] |
| val_list = [] |
|
|
| for split in os.listdir(ori_folder): |
| |
| split_path = os.path.join(ori_folder, split) |
| for folder in os.listdir(split_path): |
| |
| folder_path = os.path.join(split_path, folder) |
| for img in os.listdir(folder_path): |
| img_path = os.path.join(folder_path, img) |
| |
| if folder == 'label': |
| original_image = np.array(Image.open(img_path), dtype=np.uint8) |
| else: |
| original_image = np.asarray(Image.open(img_path).convert('RGB')) |
|
|
| for i in range(int(ori_size/crop_size)): |
| for j in range (int(ori_size/crop_size)): |
| crop = original_image[crop_size*i:crop_size*(i+1), crop_size*j:crop_size*(j+1)] |
| print(crop.shape) |
| new_name = img.split('.png')[0] + '_' + str(i) + '_' + str(j) + '.png' |
| if folder == 'label': |
| if split == 'train': train_list.append(new_name) |
| elif split == 'test': test_list.append(new_name) |
| elif split == 'val': val_list.append(new_name) |
| new_folder_path = os.path.join(cropped_folder, folder) |
| new_img_path = os.path.join(cropped_folder, folder, new_name) |
| if not os.path.exists(new_folder_path): |
| os.makedirs(new_folder_path) |
| im = Image.fromarray(crop) |
| im.save(new_img_path) |
| print('Write to : ', new_img_path) |
|
|
| |
| list_folder = os.path.join(cropped_folder, "list") |
| if not os.path.exists(list_folder): |
| os.makedirs(list_folder) |
| with open(os.path.join(list_folder, "train.txt"), "w") as f: |
| for img in train_list: |
| f.write(img) |
| if img != train_list[-1]: |
| f.write("\n") |
|
|
| with open(os.path.join(list_folder, "test.txt"), "w") as f: |
| for img in test_list: |
| f.write(img) |
| if img != test_list[-1]: |
| f.write("\n") |
|
|
| with open(os.path.join(list_folder, "val.txt"), "w") as f: |
| for img in val_list: |
| f.write(img) |
| if img != val_list[-1]: |
| f.write("\n") |
|
|
| def crop_s2looking(): |
| ''' |
| Crop S2Looking dataset in original resolution with this folder structure: |
| !!! Make sure the train, test, and val folders are named exactly like what is written below |
| -S2Looking/ |
| ├─train |
| ├─Image1 |
| ├─1.png |
| ... |
| ├─Image2 |
| ├─1.png |
| ... |
| ├─label |
| ├─1.png |
| ... |
| ├─label1 |
| ├─1.png |
| ... |
| ├─label2 |
| ├─1.png |
| ... |
| ├─test |
| ├─Image1 |
| ├─2.png |
| ... |
| ├─Image2 |
| ├─2.png |
| ... |
| ├─label |
| ├─2.png |
| ... |
| ├─label1 |
| ├─2.png |
| ... |
| ├─label2 |
| ├─2.png |
| ... |
| ├─val |
| ├─Image1 |
| ├─6.png |
| ... |
| ├─Image2 |
| ├─6.png |
| ... |
| ├─label |
| ├─6.png |
| ... |
| ├─label1 |
| ├─6.png |
| ... |
| ├─label2 |
| ├─6.png |
| ... |
| |
| |
| to 256 x 256 images with this folder structure |
| -S2Looking-Cropped |
| ├─A |
| ├─img1.png |
| ... |
| ├─B |
| ├─img1.png |
| ... |
| ├─label |
| ├─img1.png |
| ... |
| └─list |
| ├─val.txt |
| ├─test.txt |
| └─train.txt |
| ''' |
|
|
| |
| ori_folder = r"/mnt/c/Users/prisc/OneDrive - unige.it/Documents/Dataset/S2Looking/S2Looking" |
| |
| cropped_folder = r"/mnt/c/Users/prisc/OneDrive - unige.it/Documents/Dataset/S2Looking/S2Looking-Cropped" |
|
|
| folder_mapping = {'Image1' : 'A', 'Image2' : 'B', 'label' : 'label'} |
| ori_size = 1024 |
| crop_size = 256 |
|
|
| |
| train_list = [] |
| test_list = [] |
| val_list = [] |
|
|
| for split in os.listdir(ori_folder): |
| |
| split_path = os.path.join(ori_folder, split) |
| for src_folder, dest_folder in folder_mapping.items(): |
| |
| folder_path = os.path.join(split_path, src_folder) |
| for img in os.listdir(folder_path): |
| img_path = os.path.join(folder_path, img) |
| |
| if src_folder == 'label': |
| original_image = np.array(Image.open(img_path), dtype=np.uint8) |
| else: |
| original_image = np.asarray(Image.open(img_path).convert('RGB')) |
|
|
| for i in range(int(ori_size/crop_size)): |
| for j in range (int(ori_size/crop_size)): |
| crop = original_image[crop_size*i:crop_size*(i+1), crop_size*j:crop_size*(j+1)] |
| print(crop.shape) |
| new_name = img.split('.png')[0] + '_' + str(i) + '_' + str(j) + '.png' |
| if src_folder == 'label': |
| if split == 'train': train_list.append(new_name) |
| elif split == 'test': test_list.append(new_name) |
| elif split == 'val': val_list.append(new_name) |
| new_folder_path = os.path.join(cropped_folder, dest_folder) |
| new_img_path = os.path.join(cropped_folder, dest_folder, new_name) |
| if not os.path.exists(new_folder_path): |
| os.makedirs(new_folder_path) |
| im = Image.fromarray(crop) |
| im.save(new_img_path) |
| print('Write to : ', new_img_path) |
|
|
| |
| list_folder = os.path.join(cropped_folder, "list") |
| if not os.path.exists(list_folder): |
| os.makedirs(list_folder) |
| with open(os.path.join(list_folder, "train.txt"), "w") as f: |
| for img in train_list: |
| f.write(img) |
| if img != train_list[-1]: |
| f.write("\n") |
|
|
| with open(os.path.join(list_folder, "test.txt"), "w") as f: |
| for img in test_list: |
| f.write(img) |
| if img != test_list[-1]: |
| f.write("\n") |
|
|
| with open(os.path.join(list_folder, "val.txt"), "w") as f: |
| for img in val_list: |
| f.write(img) |
| if img != val_list[-1]: |
| f.write("\n") |
|
|
| def crop_whu(): |
| ''' |
| First, change the original folder structure of the whole image from this: |
| -WHU/ |
| ├─2012 |
| ├─whole_image |
| ├─test |
| ├─image |
| ├─2012_test.tif |
| ├─label |
| ├─2012_test.tif |
| ├─train |
| ├─image |
| ├─2012_train.tif |
| ├─label |
| ├─2012_train.tif |
| ├─splited_images |
| ... |
| |
| ├─2016 |
| ├─whole_image |
| ├─test |
| ├─image |
| ├─2016_test.tif |
| ├─label |
| ├─2016_test.tif |
| ├─train |
| ├─image |
| ├─2016_train.tif |
| ├─label |
| ├─2016_train.tif |
| ├─splited_images |
| ... |
| ├─change_label |
| ├─test |
| ├─change_label.tif |
| ├─train |
| ├─change_label.tif |
| |
| to this: |
| |
| -WHU/ |
| ├─2012 |
| ├─test |
| ├─2012_test.tif |
| ├─train |
| ├─2012_train.tif |
| ├─2016 |
| ├─test |
| ├─2016_test.tif |
| ├─train |
| ├─2016_train.tif |
| ├─change_label |
| ├─test |
| ├─change_label.tif |
| ├─train |
| ├─change_label.tif |
| |
| Finally, by executing the code below, it will crop the images to 256 x 256 with this folder structure: |
| -WHU-Cropped |
| ├─A |
| ├─test |
| ├─0.png |
| ... |
| ├─train |
| ├─1.png |
| ... |
| ├─B |
| ├─test |
| ├─0.png |
| ... |
| ├─train |
| ├─1.png |
| ... |
| ├─label |
| ├─test |
| ├─0.png |
| ... |
| ├─train |
| ├─1.png |
| ... |
| └─list |
| ├─val.txt |
| ├─test.txt |
| └─train.txt |
| ''' |
| |
| ori_folder = r"/mnt/c/Users/prisc/OneDrive - unige.it/Documents/Dataset/WHU/WHU" |
| |
| cropped_folder = r"/mnt/c/Users/prisc/OneDrive - unige.it/Documents/Dataset/WHU/WHU-Cropped" |
|
|
| full_size_pre_folder = os.path.join(ori_folder, '2012') |
| full_size_post_folder = os.path.join(ori_folder, '2016') |
| full_size_label_folder = os.path.join(ori_folder, 'change_label') |
| crop_label_folder = os.path.join(cropped_folder, 'label') |
| crop_pre_folder = os.path.join(cropped_folder, 'A') |
| crop_post_folder = os.path.join(cropped_folder, 'B') |
|
|
| if not os.path.exists(crop_label_folder): |
| os.makedirs(crop_label_folder) |
| if not os.path.exists(crop_pre_folder): |
| os.makedirs(crop_pre_folder) |
| if not os.path.exists(crop_post_folder): |
| os.makedirs(crop_post_folder) |
|
|
| folder = ['train', 'test'] |
| label_file = 'change_label.tif' |
|
|
| cH, cW = 256, 256 |
|
|
| for split in folder: |
| pre_file_path = os.path.join(full_size_pre_folder, split, ('2012_' + split + '.tif')) |
| post_file_path = os.path.join(full_size_post_folder, split, ('2016_' + split + '.tif')) |
| label_file_path = os.path.join(full_size_label_folder, split, label_file) |
|
|
| label = np.array(Image.open(label_file_path), dtype=np.uint8) |
| img_pre = np.asarray(Image.open(pre_file_path).convert('RGB')) |
| img_post = np.asarray(Image.open(post_file_path).convert('RGB')) |
|
|
| H, W = label.shape |
| total = 0 |
|
|
| if split == 'train': |
| total_W = W // cW + 1 |
| else: |
| total_W = W // cW |
| total_H = H // cH + 1 |
|
|
| countwh = 0 |
| counth = 0 |
| countw = 0 |
|
|
| for i in range(total_H): |
| for j in range(total_W): |
| if i == total_H - 1 and j == total_W -1: |
| countwh += 1 |
| if split == 'train': |
| crop_pre = img_pre[cH*i-6:cH*(i+1), cW*j-5:cW*(j+1)] |
| crop_post = img_post[cH*i-6:cH*(i+1), cW*j-5:cW*(j+1)] |
| crop_label = label[cH*i-6:cH*(i+1), cW*j-5:cW*(j+1)] |
| else: |
| crop_pre = img_pre[cH*i-6:cH*(i+1), cW*j:cW*(j+1)] |
| crop_post = img_post[cH*i-6:cH*(i+1), cW*j:cW*(j+1)] |
| crop_label = label[cH*i-6:cH*(i+1), cW*j:cW*(j+1)] |
| elif i == total_H - 1: |
| counth += 1 |
| crop_pre = img_pre[cH*i-6:cH*(i+1), cW*j:cW*(j+1)] |
| crop_post = img_post[cH*i-6:cH*(i+1), cW*j:cW*(j+1)] |
| crop_label = label[cH*i-6:cH*(i+1), cW*j:cW*(j+1)] |
| elif j == total_W -1: |
| countw += 1 |
| if split == 'train': |
| crop_pre = img_pre[cH*i:cH*(i+1), cW*j-5:cW*(j+1)] |
| crop_post = img_post[cH*i:cH*(i+1), cW*j-5:cW*(j+1)] |
| crop_label = label[cH*i:cH*(i+1), cW*j-5:cW*(j+1)] |
| else: |
| crop_pre = img_pre[cH*i:cH*(i+1), cW*j:cW*(j+1)] |
| crop_post = img_post[cH*i:cH*(i+1), cW*j:cW*(j+1)] |
| crop_label = label[cH*i:cH*(i+1), cW*j:cW*(j+1)] |
| else: |
| crop_pre = img_pre[cH*i:cH*(i+1), cW*j:cW*(j+1)] |
| crop_post = img_post[cH*i:cH*(i+1), cW*j:cW*(j+1)] |
| crop_label = label[cH*i:cH*(i+1), cW*j:cW*(j+1)] |
| |
| crop_label[crop_label == 1] = 255 |
|
|
| pre_H, pre_W, _ = crop_pre.shape |
| post_H, post_W, _ = crop_post.shape |
| label_H, label_W = crop_label.shape |
|
|
| print('Pre image size : ', crop_pre.shape) |
| print('Post image size : ', crop_post.shape) |
| print('Label size : ', crop_label.shape) |
|
|
| if pre_H != 256 or pre_W != 256 or post_H != 256 or post_W != 256 or label_H != 256 or label_W != 256: |
| exit(-1) |
|
|
| new_name = split + '_' + str(total) + '.png' |
| new_pre_path = os.path.join(crop_pre_folder, split) |
| new_pre_img_path = os.path.join(crop_pre_folder, split, new_name) |
| new_post_path = os.path.join(crop_post_folder, split) |
| new_post_img_path = os.path.join(crop_post_folder, split, new_name) |
| new_label_path = os.path.join(crop_label_folder, split) |
| new_label_img_path = os.path.join(crop_label_folder, split, new_name) |
| if not os.path.exists(new_pre_path): |
| os.makedirs(new_pre_path) |
| if not os.path.exists(new_post_path): |
| os.makedirs(new_post_path) |
| if not os.path.exists(new_label_path): |
| os.makedirs(new_label_path) |
|
|
| lbl = Image.fromarray(crop_label) |
| lbl.save(new_label_img_path) |
| pre = Image.fromarray(crop_pre) |
| pre.save(new_pre_img_path) |
| post = Image.fromarray(crop_post) |
| post.save(new_post_img_path) |
| total += 1 |
|
|
| print('Write to : ', new_label_img_path) |
| print('Write to : ', new_pre_img_path) |
| print('Write to : ', new_post_img_path) |
|
|
| if __name__ == "__main__": |
| |
| crop_levir() |
|
|
| |
| |
|
|
| |
| |
