STAT USER GUIDE

User Manual:

Open the PDF directly: View PDF PDF.
Page Count: 3

DownloadSTAT-USER-GUIDE
Open PDF In BrowserView PDF
STAT: Scene Text Annotation Tool
v 0.1
by Muhammet Baştan
mubastan@gmail.com

Date: 20 July 2015, Update: 01 Aug 2015, 12 Aug 2016, 15 Feb 2019

Figure 1: STAT screenshot
STAT is a simple scene text annotation tool (written in Python using PyQT4), to
select and annotate objects or text in images. The object/text selection is
either by painting over the object with the mouse, or by drawing a polygon. In
both cases, the minimum bounding box and the object mask as a bitmap
are saved on the disk. The objects can be labeled; the labels can be UTF-8 text
(tested for Turkish, but not for any other language).
Version 0.1 of the tool is developed specifically to annotate scene text regions,
but it can also be used to annotate other object categories.
The annotations are saved as text files (+ .png files for object masks). It can be
easily modified to save in xml format, if needed.

1. Installation
Required: Python 2.7, PyQT4
Developed and tested on Linux Mint 17.1.
Open a terminal in the directory of the tool and just type: python STAT.py
Then, you should see the above GUI (without any images loaded).
It is adapted from the X-ray Image Annotation Tool (XRanT), and the current

code is unfortunately not in a good state; feel free to contribute to improve it.
XRanT: https://sites.google.com/site/mubastan/research/software/xrant

2. Using STAT
The images to be annotated need to be in a single directory. Then, open this
directory to load the list of images in one of these ways:
>> File > Open Image directory
>> Ctrl + O
>> Click on Change/load button at the top right corner
The annotations will be saved in a subdirectory called “ann” under the image
directory. So, make sure you have write permissions for the image directory
otherwise the annotations will not be saved.
Planned: Changing the output directory (You can open Ann.py with a text
editor, go to line 'self.annotationDir = str(imageDir + '/ann/')' and change the
directory to anything, e.g., self.annotationDir = '/home/ali/annotations/')
The GUI has 3 main panels:
• Image panel/canvas on the left: here you select the objects with the
mouse (painting, polygon). Right click to see what operations are
available (add object, hide/show brush,...)
• Image panel in the middle: this shows the currently selected objects. You
can right click on the objects to update their label or delete them. You
can right click on the background to delete all objects, etc.
• The table on the right: shows the list of all images in the image directory,
and number of objects selected.
Hint: hover you mouse over the GUI elements to see short help texts on the
status bar.
Button next (shortcut: space): go to the next image; current image annotation
(bounding boxes + bitmaps) is saved, if anything has changed since last save.
Button previous (shortcut: Ctrl+space): go to the previous image; current
image annotation (bounding boxes + bitmaps) is saved, if anything has
changed since last save.
Button save (shortcut: Ctrl+S): save current image annotation
boxes + bitmaps).

(bounding

Text field at the bottom: shows the label of the currently selected object. You
can change the label of the currently selected object by entering the new label
and pressing Enter (or right click on the object and select 'update label').
Brush type combo box: for selecting the brush type. You can also change the
brush color by clicking 'Brush color' button.
Slider: for selecting the width of brush for painting.
Button Exit: Exits the application, by first confirming, then saving if required.

Since everything is saved on the disk, when you re-load the image directory
after you close and re-open the application, it checks the already available
annotations in the annotation directory and loads them automatically. So, you
can do annotation in multiple sessions, you can update annotations, etc.

3. Outputs
Outputs are saved to the annotation directory.
1. Object masks, as png files; one image for each object: image.0.png,
image.1.png,...
2. Label files: bounding box and label for each bounding box, to a txt file for
each image: image.png.labels.txt
Example: image.png.labels.txt
image.png
0 6 246 65 34 Text1

# objectID leftCoord topCoord W H Label

2 81 75 89 35 Text2
4 15 70 67 29 Text3
1 8 181 65 26 Text4
3 14 122 103 30 Text5
3. Only the bounding boxes into a txt file for each image
Note: this is done only when you do: File > Save all annotation bboxes
Then, this file is saved for all images.
Example: image.png.bbox.txt
/home/research/code/imant-py/text/image.png
238 293

# ImageWidth ImageHeight

6 246 65 34

# leftCoord topCoord width height

81 75 89 35
15 70 67 29
8 181 65 26
14 122 103 30
Source Exif Data: 
File Type                       : PDF
File Type Extension             : pdf
MIME Type                       : application/pdf
PDF Version                     : 1.4
Linearized                      : No
Page Count                      : 3
Language                        : en-US
Author                          : Limon Mint
Creator                         : Writer
Producer                        : LibreOffice 5.1
Create Date                     : 2019:02:15 15:45:22+08:00
EXIF Metadata provided by EXIF.tools

Navigation menu