Wrong x coordinates for bounding boxes

#2
by SouthPAW - opened

Looking further into this data, the x coords seem to be off by a random amount for every bounding box. Would love to know how to work around this or why this is caused

From what I can tell the horizontal coordinates are not off by random amount. Multiplying the horizontal coordinates by about 0.75 seems to recover the original bounding boxes. Also I have seen a similar discrepancy on other versions of FUNSD on huggingface.

The version from the FUNSD paper's first author does not have this issue.
https://guillaumejaume.github.io/FUNSD/dataset.zip

Sign up or log in to comment