Enhancement Request- Support for Unicode labels

[mineshmathew]

I wanted to assign unicode strings to my bounding boxes. I can enter my strings to the text box in UI and is being rendered correctly. But upon save it throws an error.

Traceback (most recent call last):
  File "./labelImg.py", line 945, in saveFile
    else self.saveFileDialog())
  File "./labelImg.py", line 968, in _saveFile
    if filename and self.saveLabels(filename):
  File "./labelImg.py", line 589, in saveLabels
    self.lineColor.getRgb(), self.fillColor.getRgb())
  File "libs/labelFile.py", line 46, in savePascalVocFormat
    writer.save(targetFile = filename)
  File "libs/pascal_voc_io.py", line 96, in save
    self.appendObjects(root)
  File "libs/pascal_voc_io.py", line 77, in appendObjects
    name.text = str(each_object['name'])
UnicodeEncodeError: 'ascii' codec can't encode characters in position 0-10: ordinal not in range(128)

I tried to fix it by the way the string is encoded before being written to xml. But the best I could get was a sequence of unicode points like this written to the xml in place of the actual unicode string.
<name>&#3381;&#3391;&#3349;&#3405;&#3349;&#3391;&#3370;&#3392;&#3361;&#3391;&#3375;</name>

An enhancement to support unicode strings would of great help for all who want to assign labels in languages other than English

[rflynn]

@mineshmathew try the latest version

[mineshmathew]

thank you. it works now

[mineshmathew]

@rflynn sorry to tag you again. The issue with unicode is still there.
LIke I pointed out in the first post, though the label is rendered correctly in the tool, the xml saves shows the label as a string of unicode points.

For example, for the Chinese word 中华人民共和国,
what I see in xml is this
<name>&#20013;&#21326;&#20154;&#27665;&#20849;&#21644;&#22269;</name>

[tzutalin]

@mineshmathew ,
Can you share us with the filename and the file? We can put the file to https://github.com/tzutalin/labelImg/tree/master/tests, we can test it whenever there are new changes.

[mineshmathew]

the image i used is one of the icons that comes with the code - /icons/delete.png
I am attaching the xml generated upon assigning the label 中华人民共和国 to some bounding box I drew
delete.txt

[jerinphilip]

Hello, @rflynn, @mineshmathew forwarded this to me. I took a look at your branch corresponding to this issue. The last bit was failing, due to python2 open's ascii codec issues. Since this is python2, I've worked up a solution using codecs, where it was failing.

unicode-solution.patch

[mineshmathew]

@jerinphilip thanks a lot. it works now

[tzutalin]

@jerinphilip ,
Thanks. I applied your patch to master branch.
1fb28d8

[jerinphilip]

@tzutalin: My patch alone wouldn't be enough, I'm guessing. I believe the changes @rflynn was trying to implement are also necessary. My patch is on top of @rflynn's branch corresponding to the repo.

[tzutalin]

Fix 70721ad