SVHN dataset integration

[felixdittrich92]

This PR integrates the SVHN dataset for digit localization
#587

• check boxes
• rework _get_coords

Any suggestions are very welcome 🤗

[fg-mindee]

Thanks Felix! I added a few suggestions, including switching to scipy

[codecov]

Codecov Report

Merging #634 (273598f) into main (618b0fa) will increase coverage by 0.13%.
The diff coverage is 94.87%.

@@            Coverage Diff             @@
##             main     #634      +/-   ##
==========================================
+ Coverage   96.20%   96.34%   +0.13%     
==========================================
  Files         124      125       +1     
  Lines        4668     4707      +39     
==========================================
+ Hits         4491     4535      +44     
+ Misses        177      172       -5     

Flag	Coverage Δ
unittests	96.34% <94.87%> (+0.13%)	⬆️

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

Impacted Files	Coverage Δ
doctr/datasets/synthtext.py	97.22% <ø> (ø)
doctr/datasets/svhn.py	94.73% <94.73%> (ø)
doctr/datasets/__init__.py	100.00% <100.00%> (ø)
doctr/models/recognition/predictor/pytorch.py	90.90% <0.00%> (-6.07%)	⬇️
...dels/detection/differentiable_binarization/base.py	93.88% <0.00%> (+2.77%)	⬆️
doctr/models/detection/linknet/base.py	89.10% <0.00%> (+3.96%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 618b0fa...273598f. Read the comment docs.

[fg-mindee]

Thanks, I added a snippet to replace the data unpacking part 👍

[fg-mindee]

#694 is merged 👍

I realized SVHN is processing train & test sets exactly the same way, so no need for the extra test case!

[fg-mindee]

That's some neat monkeypatching 👍
Thanks a lot!