[Fix] Resize interpolation difference CV2 and Pillow to come closer with docTR

[felixdittrich92]

This PR:

• move to Pillow for resizing to overcome the interpolation mismatch between CV2 and Pillow

[codecov-commenter]

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

Attention: Patch coverage is 97.05882% with 1 line in your changes missing coverage. Please review.

Project coverage is 97.86%. Comparing base (03a1277) to head (0f05709).

Files	Patch %	Lines
onnxtr/transforms/base.py	96.42%	1 Missing ⚠️

❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #22      +/-   ##
==========================================
+ Coverage   97.81%   97.86%   +0.05%     
==========================================
  Files          63       63              
  Lines        1920     1920              
==========================================
+ Hits         1878     1879       +1     
+ Misses         42       41       -1     

Flag	Coverage Δ
unittests	97.86% <97.05%> (+0.05%)	⬆️

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]

Overview:

Before:

(1024, 1024, 1)
(1024, 1024, 1)
Det model: db_resnet34
Mean difference between DOCTR and ONNXTR prob maps: 0.00266014
(1024, 1024, 1)
(1024, 1024, 1)
Det model: db_resnet50
Mean difference between DOCTR and ONNXTR prob maps: 0.00546016
(1024, 1024, 1)
(1024, 1024, 1)
Det model: db_mobilenet_v3_large
Mean difference between DOCTR and ONNXTR prob maps: 0.00505121
(1024, 1024, 1)
(1024, 1024, 1)
Det model: linknet_resnet18
Mean difference between DOCTR and ONNXTR prob maps: 0.00332537
(1024, 1024, 1)
(1024, 1024, 1)
Det model: linknet_resnet34
Mean difference between DOCTR and ONNXTR prob maps: 0.00293290
(1024, 1024, 1)
(1024, 1024, 1)
Det model: linknet_resnet50
Mean difference between DOCTR and ONNXTR prob maps: 0.00237311
(1024, 1024, 1)
(1024, 1024, 1)
Det model: fast_tiny
Mean difference between DOCTR and ONNXTR prob maps: 0.07307008
(1024, 1024, 1)
(1024, 1024, 1)
Det model: fast_small
Mean difference between DOCTR and ONNXTR prob maps: 0.07393986
(1024, 1024, 1)
(1024, 1024, 1)
Det model: fast_base
Mean difference between DOCTR and ONNXTR prob maps: 0.07267220

After:

(1024, 1024, 1)
(1024, 1024, 1)
Det model: db_resnet34
Mean difference between DOCTR and ONNXTR prob maps: 0.00000007
(1024, 1024, 1)
(1024, 1024, 1)
Det model: db_resnet50
Mean difference between DOCTR and ONNXTR prob maps: 0.00005615
(1024, 1024, 1)
(1024, 1024, 1)
Det model: db_mobilenet_v3_large
Mean difference between DOCTR and ONNXTR prob maps: 0.00003883
(1024, 1024, 1)
(1024, 1024, 1)
Det model: linknet_resnet18
Mean difference between DOCTR and ONNXTR prob maps: 0.00000575
(1024, 1024, 1)
(1024, 1024, 1)
Det model: linknet_resnet34
Mean difference between DOCTR and ONNXTR prob maps: 0.00002059
(1024, 1024, 1)
(1024, 1024, 1)
Det model: linknet_resnet50
Mean difference between DOCTR and ONNXTR prob maps: 0.00000474
(1024, 1024, 1)
(1024, 1024, 1)
Det model: fast_tiny
Mean difference between DOCTR and ONNXTR prob maps: 0.07032311
(1024, 1024, 1)
(1024, 1024, 1)
Det model: fast_small
Mean difference between DOCTR and ONNXTR prob maps: 0.07004898
(1024, 1024, 1)
(1024, 1024, 1)
Det model: fast_base
Mean difference between DOCTR and ONNXTR prob maps: 0.07088637