alek-dr/pascalvoc

Python utility to work with PascalVoc annotation format

pascal

Python utility to work with PascalVoc annotation format

Image examples from PascalVoc2007 dataset

Code Example

import json
from pathlib import Path

from PIL import Image

from pascal import annotation_from_xml
from pascal.utils import save_xml

ds_path = Path("/home/VOCtest_06-Nov-2007/VOCdevkit/VOC2007")

img_src = ds_path / "JPEGImages"
ann_src = ds_path / "Annotations"

out_labelme = ds_path / "converted_labelme"

attr_type_spec = {"truncated": bool, "difficult": bool}

label_map = {"car": 1, "dog": 0, "person": 2, "train": 3}

if __name__ == "__main__":
    for file in img_src.glob("*.jpg"):
        # Get annotation file path
        ann_file = (ann_src / file.name).with_suffix(".xml")
        # Make pascal annotation object
        ann = annotation_from_xml(ann_file, attr_type_spec)
        print(ann)
        # Save to xml file (same as ann_file)
        xml = ann.to_xml()
        out_xml_name = file.with_suffix(".xml").name
        save_xml(out_xml_name, xml)
        # Save yolo annotation
        yolo_ann = ann.to_yolo(label_map)
        out_yolo_name = file.with_suffix(".txt").name
        with open(out_yolo_name, "w") as f:
            f.write(yolo_ann)
        # Convert to labelme and save json file
        res = ann.to_labelme(file, save_img_data=False)
        with open((out_labelme / file.name).with_suffix(".json"), "w") as f:
            json.dump(res, f, indent=2)
        # Draw objects
        img = Image.open(file)
        draw_img = ann.draw_boxes(img)
        draw_img.show()

Visualization example:

draw_img = ann.draw_boxes(img)
draw_img.show()

Installation

From source

python setup.py install

Using pip

pip install pascal-voc

pascalvoc's Issues

negative bndbox values are loaded as string

When loading annotations from xml with the annotation_from_xml function the library loads negative bndbox values as type. This seems wrong and causes issues when using other functions like draw_boxes.

The xml shouldn't have negative values but it happens with some labeling tools. In my case i used https://www.makesense.ai/

Example voc_xml:

<annotation>
	<folder>my-project-name</folder>
	<filename>test.jpg</filename>
	<path>/my-project-name/test.jpg</path>
	<source>
		<database>Unspecified</database>
	</source>
	<size>
		<width>355</width>
		<height>539</height>
		<depth>3</depth>
	</size>
	<object>
		<name>cat1</name>
		<pose>Unspecified</pose>
		<truncated>0</truncated>
		<difficult>0</difficult>
		<bndbox>
			<xmin>-2</xmin>
			<ymin>-3</ymin>
			<xmax>355</xmax>
			<ymax>333</ymax>
		</bndbox>
	</object>
	<object>
		<name>cat1</name>
		<pose>Unspecified</pose>
		<truncated>0</truncated>
		<difficult>0</difficult>
		<bndbox>
			<xmin>230</xmin>
			<ymin>441</ymin>
			<xmax>336</xmax>
			<ymax>525</ymax>
		</bndbox>
	</object>
</annotation>