Images Annotation Programme

Free Online web tool to annotate images, output format is a list of xml files (Pascal VOC xml format). This image labelling application will help you creating a learning base for image recognition.

Discover an example: http://bipbipavertisseur.alwaysdata.net/image_annotations_v2.0_d1

Customization

1. Configuration

To customize the directories used, edit the PHP file inc/configuration.php

<?php
# Image path to be used in the HTML client
$IMAGE_WEB_DIR = "data/images";

# Image path for internal PHP use
$IMAGE_ROOT_DIR  = "../data/images";
$ANNOTATIONS_DIR = "../data/annotations";

# Not annotated image 80% to be presented to user
$ratio_new_old = 80;
?>

2. Images

Images to be annotated are located in data/images

3. List of classes

The list of classes can be customized in the file resources/list_of_tags.json

[
	{"name": "Long Beak Bird", "icon": "resources/tag_examples/long_beak.jpg"},
	{"name": "Eagle", "icon": "resources/tag_examples/eagle.jpg"},
	{"name": "Parrot", "icon": "resources/tag_examples/parrot.jpg"},	
	{"name": "Baby Bird", "icon": "resources/tag_examples/baby_bird.jpg"}
]

The result is quite cool !

4. Annotations Target directory

Each image will generate one XML file in the directory data/annotations

Output as Pascal VOC xml files

This format is a standard and can be easily read from Tensorflow Object Detection API

<?xml version="1.0"?>
<annotation>
  <folder>images</folder>
  <filename>pexels-photo-60091.jpg</filename>
  <path/>
  <source>
    <database>Unknown</database>
  </source>
  <size>
     <width>1125</width>
     <height>750</height>
     <depth>3</depth>
  </size>
  <segmented>0</segmented>
  <object>
    <name>Bird</name>
    <pose>Unspecified</pose>
    <truncated>0</truncated>
    <difficult>0</difficult>
    <bndbox>
      <xmin>488</xmin>
      <ymin>245.5</ymin>
      <xmax>674</xmax>
      <ymax>601.5</ymax>
    </bndbox>
  </object>
</annotation>

Generate TFRecord file format

# From shell/ (by default config)
python convert_tag_json_to_pbtxt.py

file shell/label_map.pbtxt should be generate

# From shell/ (by default config)
# note:tensorflow should be installed success
python convert_voc_to_rf_record.py

file shell/train.record should be generate

Contributions

Many thanks to the contributors of these useful libraries:

• jQuery-select-areas available on GitHub
• EasyAutocomplete available on GitHub

I modified some pieces of code to adapt the features to my needs.

License

Code released under the MIT license.