Allow XML dataset generalization (#5943)

mmdet/datasets/xml_style.py

@@ -18,11 +18,19 @@ class XMLDataset(CustomDataset):
         min_size (int | float, optional): The minimum size of bounding
             boxes in the images. If the size of a bounding box is less than
             ``min_size``, it would be add to ignored field.
+        img_subdir (str): Subdir where images are stored. Default: JPEGImages.
+        ann_subdir (str): Subdir where annotations are. Default: Annotations.
     """
 
-    def __init__(self, min_size=None, **kwargs):
+    def __init__(self,
+                 min_size=None,
+                 img_subdir='JPEGImages',
+                 ann_subdir='Annotations',
+                 **kwargs):
         assert self.CLASSES or kwargs.get(
             'classes', None), 'CLASSES in `XMLDataset` can not be None.'
+        self.img_subdir = img_subdir
+        self.ann_subdir = ann_subdir
         super(XMLDataset, self).__init__(**kwargs)
         self.cat2label = {cat: i for i, cat in enumerate(self.CLASSES)}
         self.min_size = min_size
@@ -40,8 +48,8 @@ class XMLDataset(CustomDataset):
         data_infos = []
         img_ids = mmcv.list_from_file(ann_file)
         for img_id in img_ids:
-            filename = f'JPEGImages/{img_id}.jpg'
-            xml_path = osp.join(self.img_prefix, 'Annotations',
+            filename = osp.join(self.img_subdir, f'{img_id}.jpg')
+            xml_path = osp.join(self.img_prefix, self.ann_subdir,
                                 f'{img_id}.xml')
             tree = ET.parse(xml_path)
             root = tree.getroot()
@@ -50,8 +58,7 @@ class XMLDataset(CustomDataset):
                 width = int(size.find('width').text)
                 height = int(size.find('height').text)
             else:
-                img_path = osp.join(self.img_prefix, 'JPEGImages',
-                                    '{}.jpg'.format(img_id))
+                img_path = osp.join(self.img_prefix, filename)
                 img = Image.open(img_path)
                 width, height = img.size
             data_infos.append(
@@ -67,7 +74,7 @@ class XMLDataset(CustomDataset):
                 continue
             if self.filter_empty_gt:
                 img_id = img_info['id']
-                xml_path = osp.join(self.img_prefix, 'Annotations',
+                xml_path = osp.join(self.img_prefix, self.ann_subdir,
                                     f'{img_id}.xml')
                 tree = ET.parse(xml_path)
                 root = tree.getroot()
@@ -91,7 +98,7 @@ class XMLDataset(CustomDataset):
         """
 
         img_id = self.data_infos[idx]['id']
-        xml_path = osp.join(self.img_prefix, 'Annotations', f'{img_id}.xml')
+        xml_path = osp.join(self.img_prefix, self.ann_subdir, f'{img_id}.xml')
         tree = ET.parse(xml_path)
         root = tree.getroot()
         bboxes = []
@@ -158,7 +165,7 @@ class XMLDataset(CustomDataset):
 
         cat_ids = []
         img_id = self.data_infos[idx]['id']
-        xml_path = osp.join(self.img_prefix, 'Annotations', f'{img_id}.xml')
+        xml_path = osp.join(self.img_prefix, self.ann_subdir, f'{img_id}.xml')
         tree = ET.parse(xml_path)
         root = tree.getroot()
         for obj in root.findall('object'):

tests/data/custom_dataset/images/000001.xml

@@ -0,0 +1,44 @@
+<annotation>
+	<folder>VOC2007</folder>
+	<filename>000001.jpg</filename>
+	<source>
+		<database>The VOC2007 Database</database>
+		<annotation>PASCAL VOC2007</annotation>
+		<image>flickr</image>
+		<flickrid>341012865</flickrid>
+	</source>
+	<owner>
+		<flickrid>Fried Camels</flickrid>
+		<name>Jinky the Fruit Bat</name>
+	</owner>
+	<size>
+		<width>353</width>
+		<height>500</height>
+		<depth>3</depth>
+	</size>
+	<segmented>0</segmented>
+	<object>
+		<name>dog</name>
+		<pose>Left</pose>
+		<truncated>1</truncated>
+		<difficult>0</difficult>
+		<bndbox>
+			<xmin>48</xmin>
+			<ymin>240</ymin>
+			<xmax>195</xmax>
+			<ymax>371</ymax>
+		</bndbox>
+	</object>
+	<object>
+		<name>person</name>
+		<pose>Left</pose>
+		<truncated>1</truncated>
+		<difficult>0</difficult>
+		<bndbox>
+			<xmin>8</xmin>
+			<ymin>12</ymin>
+			<xmax>352</xmax>
+			<ymax>498</ymax>
+		</bndbox>
+	</object>
+</annotation>

tests/data/custom_dataset/test.txt

@@ -0,0 +1 @@
+000001

tests/data/custom_dataset/trainval.txt

@@ -0,0 +1 @@
+000001

tests/test_data/test_datasets/test_custom_dataset.py

@@ -1,4 +1,6 @@
 # Copyright (c) OpenMMLab. All rights reserved.
+import os
+import unittest
 from unittest.mock import MagicMock, patch
 
 import pytest
@@ -87,3 +89,51 @@ def test_custom_classes_override_default(dataset):
     assert custom_dataset.CLASSES != original_classes
     assert custom_dataset.CLASSES == ['bus', 'car']
     print(custom_dataset)
+
+
+class CustomDatasetTests(unittest.TestCase):
+
+    def setUp(self):
+        super().setUp()
+        self.data_dir = os.path.join(
+            os.path.dirname(os.path.dirname(os.path.dirname(__file__))),
+            'data')
+        self.dataset_class = DATASETS.get('XMLDataset')
+
+    def test_data_infos__default_db_directories(self):
+        """Test correct data read having a Pacal-VOC directory structure."""
+        test_dataset_root = os.path.join(self.data_dir, 'VOCdevkit', 'VOC2007')
+        custom_ds = self.dataset_class(
+            data_root=test_dataset_root,
+            ann_file=os.path.join(test_dataset_root, 'ImageSets', 'Main',
+                                  'trainval.txt'),
+            pipeline=[],
+            classes=('person', 'dog'),
+            test_mode=True)
+
+        self.assertListEqual([{
+            'id': '000001',
+            'filename': 'JPEGImages/000001.jpg',
+            'width': 353,
+            'height': 500
+        }], custom_ds.data_infos)
+
+    def test_data_infos__overridden_db_subdirectories(self):
+        """Test correct data read having a customized directory structure."""
+        test_dataset_root = os.path.join(self.data_dir, 'custom_dataset')
+        custom_ds = self.dataset_class(
+            data_root=test_dataset_root,
+            ann_file=os.path.join(test_dataset_root, 'trainval.txt'),
+            pipeline=[],
+            classes=('person', 'dog'),
+            test_mode=True,
+            img_prefix='',
+            img_subdir='images',
+            ann_subdir='images')
+
+        self.assertListEqual([{
+            'id': '000001',
+            'filename': 'images/000001.jpg',
+            'width': 353,
+            'height': 500
+        }], custom_ds.data_infos)