Layout blocks detected by models do not match blocks detected by Tesseract

[maxycn]

Describe the bug
Left:
is the result from
model = lp.EfficientDetLayoutModel("lp://PubLayNet/tf_efficientdet_d0/config")
layout = model.detect(img)
lp.draw_box(img, layout)

Right:
is the result from pytesseract.image_to_data()

Clearly the left one is not right. Is there a way to fix it?