feat: add pytorch ckpts for crnn & mobilenet_v3_large

[charlesmindee]

This PR adds pytorch ckpts for crnn_vgg_16_bn and mobilenet_v3_large, here is the benchmark performed with the ckpts:

FUNSD
Text Detection - Recall: 80.98%, Precision: 85.28%, Mean IoU: 68.91%
Text Recognition - Accuracy: 85.85% (unicase: 86.76%)
OCR - Recall: 67.90% (unicase: 68.52%), Precision: 71.51% (unicase: 72.17%), Mean IoU: 68.91%

CORD
Text Detection - Recall: 80.48%, Precision: 66.62%, Mean IoU: 56.98%
Text Recognition - Accuracy: 92.53% (unicase: 92.92%)
OCR - Recall: 70.80% (unicase: 71.12%), Precision: 58.60% (unicase: 58.86%), Mean IoU: 56.98%

RECEIPTS
Text Detection - Recall: 84.70%, Precision: 85.62%, Mean IoU: 72.19%
Text Recognition - Accuracy: 91.70% (unicase: 92.45%)
OCR - Recall: 76.87% (unicase: 77.51%), Precision: 77.70% (unicase: 78.34%), Mean IoU: 72.19%

IDS
Text Detection - Recall: 66.32%, Precision: 59.12%, Mean IoU: 48.83%
Text Recognition - Accuracy: 64.81% (unicase: 67.50%)
OCR - Recall: 44.58% (unicase: 46.75%), Precision: 39.74% (unicase: 41.67%), Mean IoU: 48.83%

INVOICES
Text Detection - Recall: 69.46%, Precision: 72.85%, Mean IoU: 61.74%
Text Recognition - Accuracy: 90.12% (unicase: 91.59%)
OCR - Recall: 64.03% (unicase: 65.06%), Precision: 67.15% (unicase: 68.23%), Mean IoU: 61.74%

TAX FORM US
Text Detection - Recall: 80.74%, Precision: 92.92%, Mean IoU: 70.90%
Text Recognition - Accuracy: 84.30% (unicase: 84.83%)
OCR - Recall: 78.21% (unicase: 78.58%), Precision: 90.02% (unicase: 90.45%), Mean IoU: 70.90%

[fg-mindee]

Thanks! There are some rather significant differences to be noticed compared to their TF counterparts 🤔

[codecov]

Codecov Report

Merging #487 (69e267f) into main (5d1073f) will not change coverage.
The diff coverage is n/a.

@@           Coverage Diff           @@
##             main     #487   +/-   ##
=======================================
  Coverage   94.99%   94.99%           
=======================================
  Files         108      108           
  Lines        4177     4177           
=======================================
  Hits         3968     3968           
  Misses        209      209           

Flag	Coverage Δ
unittests	94.99% <ø> (ø)

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

Impacted Files	Coverage Δ
...s/detection/differentiable_binarization/pytorch.py	97.43% <ø> (ø)
doctr/models/recognition/crnn/pytorch.py	98.90% <ø> (ø)

---

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 5d1073f...69e267f. Read the comment docs.