No bbox coordinates
#1
by
TechNetiums
- opened
When using VLLM backend of this model , the output text only contains references to images like but no bouding boxes coordinates
i think it can happen for some specific images, but it would be weird if it was for all of them? have you tried some other images?
it happens with most of the images I converted from pdf to png using the pdf2image (200DPI) . It happens especially to pdf that have slightly lower resolution.
But indeed with pdf from other datasets this issue is less frequent
how to fix this? i cant see any bounding boxes even on the official demo