ds4sd

/

SmolDocling-256M-preview

like 1.14k

Follow

Docling 521

@deewalia20 it's x1, y1,x2,y2, with top-left is origin.
Each Doctag loc_ coordinate designates proportion from page image width / height in the range from 0..500

Thanks I got it now.
For a given Doctag location info: <loc_proportional_x1><loc_proportional_y1><loc_x2><loc_y2>, use the below formula where image_width and image_height are from source image.
Then use opencv or pillow to plot bounding box.

To convert these proportional coordinates to actual pixel values, you can use the following formulas:

$$\text{actual\_x1}=\left(\frac{\text{proportional\_x1}}{500}\right)\times \text{image\_width}\backslash text\{actual\backslash \_x1\}\; =\; \backslash left(\; \backslash frac\{\backslash text\{proportional\backslash \_x1\}\}\{500\}\; \backslash right)\; \backslash times\; \backslash text\{image\backslash \_width\}$$

$$\text{actual\_y1}=\left(\frac{\text{proportional\_y1}}{500}\right)\times \text{image\_height}\backslash text\{actual\backslash \_y1\}\; =\; \backslash left(\; \backslash frac\{\backslash text\{proportional\backslash \_y1\}\}\{500\}\; \backslash right)\; \backslash times\; \backslash text\{image\backslash \_height\}$$