Labelimg: All bounding box values are 0 when using YOLO as output format

[daerty0153]

Building from source to use YOLO output format. I modified the data/predefined_classes.txt to contain only a single class "person".
VOC seems to be running fine with the following output:

<annotation>
    <folder>target</folder>
    <filename>xxx.jpg</filename>
    <path>xxx.jpg</path>
    <source>
        <database>Unknown</database>
    </source>
    <size>
        <width>1080</width>
        <height>1440</height>
        <depth>3</depth>
    </size>
    <segmented>0</segmented>
    <object>
        <name>person</name>
        <pose>Unspecified</pose>
        <truncated>0</truncated>
        <difficult>0</difficult>
        <bndbox>
            <xmin>204</xmin>
            <ymin>445</ymin>
            <xmax>800</xmax>
            <ymax>1355</ymax>
        </bndbox>
    </object>
    <object>
        <name>person</name>
        <pose>Unspecified</pose>
        <truncated>0</truncated>
        <difficult>0</difficult>
        <bndbox>
            <xmin>262</xmin>
            <ymin>442</ymin>
            <xmax>859</xmax>
            <ymax>1347</ymax>
        </bndbox>
    </object>
</annotation>

But when I try to save the annotation in YOLO format, the generated file is this:

0 0.000000 0.000000 0.000000 0.000000
0 0.000000 0.000000 0.000000 0.000000

Once I save in the YOLO format and do next and come back, the bounding box also resets to (0,0) in GUI as well.

• OS: Ubuntu16.04
• PyQt version: 4

[hkfreenet]

i got the exactly same issue. Please help which step i missed. Thanks!

[cansik]

Yep, same here on MacOSX, PyQt 5

[toonvh]

same problem here

[isVoid]

@daerty0153 I was not able to reproduce your problem on my machine with your data. Found a 1080x1440 image online and pasted your label, switched to yolo and save (you need to move the handle by a bit before you can actually save). This is what I have:

0 0.464815 0.625000 0.551852 0.631944
0 0.518519 0.617014 0.551852 0.636806

afaik labelimg has specific way handling folder structures. Can you provide your folder layout? Also detail your steps a bit more if possible.

Michael

[rosyapril]

I had exactly same problem here. Tested with Python 2 + Qt4, Ubuntu 16.04. All saved values in YOLO format are 0. Even with the windows v1.6.1, all values are 0.

But it performances alright with VOC data format.

[isVoid]

@rosyapril Can you provide your test case?

[rosyapril]

@isVoid when i labeled some objects, in terminal, below information will pop out.

GLib-GIO-Message: Using the 'memory' GSettings backend. Your settings will not be saved or shared with other applications.
(u'/home/rosy/Documents/d409_2/frame0002.txt', u'/home/rosy/Documents/d409_2/classes.txt')
['Ball', 'Goalpost', 'Opponent']
[('Ball', [(0, 0), (0, 0), (0, 0), (0, 0)], None, None, False)]
Img: /home/rosy/Documents/UTRD/dataset0409/d409_2/frame0002.jpg -> Its txt: /home/rosy/Documents/UTRD/dataset0409/d409_2/frame0002.txt
(0, 0, 0, -1, -1)
(0, 0, 0, 0, 0)
[u'Ball', u'Goalpost', u'Opponent']

In file:
0 0.000000 0.000000 -1.000000 -1.000000
0 0.000000 0.000000 0.000000 0.000000

[isVoid]

@rosyapril Not able to replicate with given DIR structure and class type, I need you to provide more detailed information. Can you provide:

System Environment
Steps to reproduce
Visual Proof (screenshots, videos, text) etc.

Thanks.

[garry1ng]

@rosyapril I had the same problem with you, and after doing float type casting in YOLO output format as the commit stated below, I found it worked well in both VOC and YOLO output format.
https://github.com/garry1ng/labelImg/commit/7d4ca92d4bf24445640b60a76024b4395f66ed41
I have pulled a request and you can test it.

[vdalv]

Looks like this bug has been fixed, @tzutalin can we close this issue?