search

YuzheZhang-1999 / DiffTSR

[CVPR2024] Diffusion-based Blind Text Image Super-Resolution (Official)
63 stars 4 forks source link

Should the size of input img keep (512, 128)? #5

Closed uhSuiL closed 2 months ago

uhSuiL commented 2 months ago

If so, is there any suggested strategy to process img of abitrary size (like 300x300)?

YuzheZhang-1999 commented 2 months ago

This project is currently only applicable to text images with single-line text, and the input size is limited to 128*512 patches. You can detect the text line image from the original image through the PaddleOCR or other text detection models, then crop and resize the patches to 128 x 512.

uhSuiL commented 2 months ago

Thanks for your reply and appreciate your work