Pascal VOC XML Format is incomplete

[Mast3rChief]

Hi,

thanks for the great app. I noticed one issue.

When I try to use the annotation via the "Pascal VOC XML" format I noticed that it is incomplete and there is no way of adding the missing tags/fields automatically.

This is what I would be expecting

<annotation>
    <folder></folder>
    <filename>000001.jpg</filename>
    <path>000001.jpg</path>
    <size>
        <width>500</width>
        <height>375</height>
        <depth>3</depth>
    </size>
    <segmented>0</segmented>
    <object>
        <name>item</name>
        <pose>Unspecified</pose>
        <truncated>0</truncated>
        <difficult>0</difficult>
        <occluded>0</occluded>
        <bndbox>
            <xmin>179</xmin>
            <xmax>231</xmax>
            <ymin>85</ymin>
            <ymax>144</ymax>
        </bndbox>
    </object>
</annotation>

This is the actual result

<annotation>
    <filename>000001.jpg</filename>
    <size>
        <width>500</width>
        <height>375</height>
        <depth>3</depth>
    </size>
    <object>
        <name>item</name>
        <difficult>0</difficult>
        <bndbox>
            <xmin>179</xmin>
            <xmax>231</xmax>
            <ymin>85</ymin>
            <ymax>144</ymax>
        </bndbox>
    </object>
</annotation>

Especially the files <pose>Unspecified</pose>, <truncated>0</truncated> and <occluded>0</occluded> are important for me.

[ryouchinsa]

Thanks for writing the issue.

For example, the difficult value is used in the DOTA format. https://captain-whu.github.io/DOTA/dataset.html

x1, y1, x2, y2, x3, y3, x4, y4, category, difficult
x1, y1, x2, y2, x3, y3, x4, y4, category, difficult
...

Could you tell us in which use-case do you need those pose, truncated, and occluded values?

[Mast3rChief]

I'm working with the road sign data set

https://www.kaggle.com/datasets/andrewmvd/road-sign-detection

For this the additional fields are required

<annotation>
    <folder>images</folder>
    <filename>road0.png</filename>
    <size>
        <width>267</width>
        <height>400</height>
        <depth>3</depth>
    </size>
    <segmented>0</segmented>
    <object>
        <name>trafficlight</name>
        <pose>Unspecified</pose>
        <truncated>0</truncated>
        <occluded>0</occluded>
        <difficult>0</difficult>
        <bndbox>
            <xmin>98</xmin>
            <ymin>62</ymin>
            <xmax>208</xmax>
            <ymax>232</ymax>
        </bndbox>
    </object>
</annotation>

Especially because my training pipeline requires the missing fields it takes some effort to add all missing parts.

[ryouchinsa]

Thanks for your detailed feedback.

We will update so that you can edit truncated, occluded, and difficult values and save them in the PASCAL VOC xml file. When the new update is released, we will let you know.

<annotation>
    <folder>test_all_1</folder>
    <filename>tuce-311434-unsplash2.jpg</filename>
    <size>
        <width>2313</width>
        <height>3000</height>
        <depth>3</depth>
    </size>
    <object>
        <name>sneakers-converse-yellow</name>
        <pose>Unspecified</pose>
        <truncated>0</truncated>
        <occluded>0</occluded>
        <difficult>1</difficult>
        <bndbox>
            <xmin>1096.526123</xmin>
            <ymin>389.870392</ymin>
            <xmax>1343.419312</xmax>
            <ymax>621.793030</ymax>
        </bndbox>
    </object>
</annotation>

[Mast3rChief]

Perfect Thanks a lot.

One thing. The <pose> filed will be available too, right? From your output it looks like it but it is not visible in the settings menu.

[ryouchinsa]

Thanks for your detailed feedback. It is because you can edit truncated, occluded, and difficult, but you cannot edit the pose tag. The pose tag is always saved as Unspecified. Please let us know if you need to edit the pose tag.

[Mast3rChief]

Sounds good. No I don’t need to edit the pose tag.

[ryouchinsa]

The new update version 74 was released. Please give us your feedback.

[ryouchinsa]

• Fixed the problem of changing brightness/contrast for png files which have 4 channels in the Road Sign Detection dataset. https://www.kaggle.com/datasets/andrewmvd/road-sign-detection
• When import, for example, a COCO JSON file, default values of pose, truncated, occluded, and difficult are saved in the PASCAL VOC xml files.

The new update version 76 was released.

[ryouchinsa]

Currently If you need our support to fix this problem, please let us know. If you have any other requests, please let us know.

[Mast3rChief]

Works fine now. Thanks