Fixed assume_straight_pages for custom models

[Fabioomega]

When loading a custom model using ocr_predictor with assume_straight_pages = False, the preprocessor will be initialized with assume_straight_pages = True, resulting in a crash. This behavior happens in any custom detection models.
Example with the default db_resnet50:

from doctr.models import ocr_predictor, db_resnet50
from torch import load
from cv2 import imread

det_arch = db_resnet50(pretrained=False, pretrained_backbone=False)
params = load("db_resnet50-79bd7d70.pt", map_location="cpu")
det_arch.load_state_dict(params)

reader = ocr_predictor(det_arch=det_arch, assume_straight_pages=False)

img = imread("<img>")
print(
    reader([img])
)  # => IndexError: too many indices for array: array is 2-dimensional, but 3 were indexed

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 96.37%. Comparing base (1cea7d8) to head (6ab97f2).

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1681      +/-   ##
==========================================
+ Coverage   96.35%   96.37%   +0.01%     
==========================================
  Files         164      164              
  Lines        7773     7774       +1     
==========================================
+ Hits         7490     7492       +2     
+ Misses        283      282       -1     

Flag	Coverage Δ
unittests	96.37% <100.00%> (+0.01%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[felixdittrich92]

Thanks for the fix @Fabioomega 👍