search

ViTAE-Transformer / DeepSolo

The official repo for [CVPR'23] "DeepSolo: Let Transformer Decoder with Explicit Points Solo for Text Spotting" & [ArXiv'23] "DeepSolo++: Let Transformer Decoder with Explicit Points Solo for Multilingual Text Spotting"
Other
250 stars 34 forks source link

Datasets conversion tool #4

Closed linxid closed 1 year ago

linxid commented 1 year ago

This is a great job, Can you provide dataset conversion script, thanks.

ymy-k commented 1 year ago

Overall, 8 bezier control points for each text instance are required, you can refer to the link in ABCNet to generate Bezier-curve. Note that, please make sure the bbox is correct (there is a issue when processing the bbox in the provided code). The conversion for sampling points already lies in here.

linxid commented 1 year ago

I have an irregularly shaped marker that starts at the top of the text box instead of the top left corner. How can I generate Bezier-curve in this case.

ymy-k commented 1 year ago

If the text is not highly curved, it doesn't matter to sample points on the axis-aligned bbox as boundary gt. Otherwise, it seems the shape of text cannot be accurately represented.

RazinAleksandr commented 1 year ago

Can you provide app which you used for data markup, thanks?