How to train my own dataset to recognize text

[JcmeLs]

Hi,

I want to use SSD to recognize text in a particular scene. I made the data set as PASCAL. The object does not know where to write the 'text', or the exact text value.

E.g

        <object>
        <name> text </name>
        <pose> Unspecified </pose>
        <truncated>0</truncated>
        <difficult>0</difficult>
        <bndbox>
            <xmin>92</xmin>
            <ymin>72</ymin>
            <xmax>305</xmax>
            <ymax>473</ymax>
        </bndbox>
    </object>

Or:

        <object>
        <name> DEMO123456789 </name>
        <pose> Unspecified </pose>
        <truncated>0</truncated>
        <difficult>0</difficult>
        <bndbox>
            <xmin>92</xmin>
            <ymin>72</ymin>
            <xmax>305</xmax>
            <ymax>473</ymax>
        </bndbox>
    </object>

End-to-end recognition can directly identify accurate text values. So...I do not know what to choose, because the data set is too time-consuming, so ask the first to do.

Looking forward to your reply. Thanks

[JcmeLs]

I create my dataset like this:

<annotation>
        <folder>STDATA</folder>
        <filename>0630selection1.JPG</filename>
        <source>
            <database>Sense Text Database</database>
            <annotation>sense text 2017</annotation>
            <image>flickr</image>
            <flickrid>001</flickrid>
        </source>
        <size>
            <width>500</width>
            <height>375</height>
            <depth>3</depth>
        </size>
        <segmented>0</segmented>
        <object>
            <name>text</name>
            <pose>Unspecified</pose>
            <truncated>0</truncated>
            <difficult>0</difficult>
            <bndbox>
                <xmin>167</xmin>
                <ymin>21</ymin>
                <xmax>325</xmax>
                <ymax>59</ymax>
            </bndbox>
        </object>
    </annotation>

And I fine-turning it.But I can't get the results what I want.