Create xml_to_segmask.py

pre_process/MoNuSeg/xml_to_segmask.py

@@ -0,0 +1,110 @@
+import xml.etree.ElementTree as ET
+import numpy as np
+import skimage.draw
+import os
+import glob
+import cv2
+from PIL import Image
+from scipy.io import savemat
+
+
+def binary_mask_from_xml_file(xml_file_path, image_shape=(1000, 1000)):
+    tree = ET.parse(xml_file_path)
+    root = tree.getroot()
+
+    def vertex_element_to_tuple(vertex_element):
+        col = float(vertex_element.get('X'))
+        row = float(vertex_element.get('Y'))
+        return round(row), round(col)
+
+    mask = np.zeros(image_shape, dtype=np.uint8)
+    iii = 1
+    for region in root.iter('Region'):
+        vertices = map(vertex_element_to_tuple, region.iter('Vertex'))
+        rows, cols = np.array(list(zip(*vertices)))
+
+        rows[rows >= 1000] = 999
+        cols[cols >= 1000] = 999       
+        rr, cc = skimage.draw.polygon(rows, cols, mask.shape)
+
+        '''
+        # To add the nuclear boundary as a separate class
+        mask[rr, cc] = 2
+        mask[rows, cols] = 1
+        '''
+        mask[rr, cc] = 1
+
+        iii += 1
+
+    return mask
+
+# TRAIN SET
+folder1 = './MoNuSeg_train/Annotations/*'  # Annotations Folder
+folder2 = './MoNuSeg_train/Tissue Images'  # Images Folder
+
+# Train Output Folders
+labelled_images = './data/train_images'
+os.makedirs(labelled_images, exist_ok=True)
+labels = './data/train_masks'
+os.makedirs(labels, exist_ok=True)
+
+labels_list = []
+IMGS = glob.glob(folder1)
+indexing = 0
+for mask_path in IMGS:
+    print(mask_path)
+    img_path = os.path.join(folder2, mask_path.split('/')[-1].split('.xml')[0] + '.tif')
+
+    image = Image.open(img_path)
+    image = np.array(image)
+    save_path1 = os.path.join(labelled_images, 'image_' + str(indexing) + '.jpg')
+    cv2.imwrite(save_path1, cv2.cvtColor(image, cv2.COLOR_RGB2BGR))
+
+    image_label = binary_mask_from_xml_file(mask_path)
+    save_path2 = os.path.join(labels, 'mask_' + str(indexing) + '.jpg')
+    cv2.imwrite(save_path2, image_label)
+    labels_list.append(image_label)
+
+    indexing += 1
+
+
+masks = np.array(labels_list)
+path = os.path.join('./data', 'train_masks.mat')
+mdic = {"data": masks, "label": "labels"}
+savemat(path, mdic)
+
+
+# TEST SET
+folder1 = './MoNuSeg_test/Annotations/*'  # Annotations Folder
+folder2 = './MoNuSeg_test/Tissue Images'  # Images Folder
+
+# Test Output Folders
+labelled_images = './data/test_images'
+os.makedirs(labelled_images, exist_ok=True)
+labels = './data/test_masks'
+os.makedirs(labels, exist_ok=True)
+
+labels_list = []
+IMGS = glob.glob(folder1)
+indexing = 0
+for mask_path in IMGS:
+    print(mask_path)
+    img_path = os.path.join(folder2, mask_path.split('/')[-1].split('.xml')[0] + '.tif')
+
+    image = Image.open(img_path)
+    image = np.array(image)
+    save_path1 = os.path.join(labelled_images, 'image_' + str(indexing) + '.jpg')
+    cv2.imwrite(save_path1, cv2.cvtColor(image, cv2.COLOR_RGB2BGR))
+
+    image_label = binary_mask_from_xml_file(mask_path)
+    save_path2 = os.path.join(labels, 'mask_' + str(indexing) + '.jpg')
+    cv2.imwrite(save_path2, image_label)
+    labels_list.append(image_label)
+
+    indexing += 1
+
+masks = np.array(labels_list)
+path = os.path.join('./data', 'test_masks.mat')
+mdic = {"data": masks, "label": "labels"}
+savemat(path, mdic)
+