| """ | |
| This script checks unique values in label | |
| """ | |
| from PIL import Image | |
| import numpy as np | |
| import glob | |
| import pdb | |
| import os | |
| from tqdm import tqdm | |
| lbl_type_list = ['train', 'val', 'test_indomain', 'test_outdomain'] | |
| for lbl_type in lbl_type_list: | |
| image_dir = '../data/'+lbl_type+'/image' | |
| label_dir = '../data/'+lbl_type+'/label' | |
| for lbl_file in tqdm(glob.glob(label_dir + '/*.png')): | |
| fname = os.path.basename(lbl_file).split('.')[0] | |
| lbl = np.array(Image.open(lbl_file)) | |
| img_file = os.path.join(image_dir, fname + '.jpg') | |
| img = np.array(Image.open(img_file)) | |
| unique_vals = np.unique(lbl) | |
| if min(unique_vals) < 0 or max(unique_vals) > 8: | |
| print('label out of bound!') | |
| if lbl.shape[0] != img.shape[0] or lbl.shape[1] != img.shape[1]: | |
| print('size mismatch!') | |
| if len(lbl.shape) != 2 or len(img.shape) != 3: | |
| print('chanel mismatch!') | |
| print(fname, lbl.shape, img.shape) | |
| os.remove(img_file) | |
| os.remove(lbl_file) | |