[detection] move padding removal directly to detection

[felixdittrich92]

This PR:

• make padding removal part of the detection predictor
• make the detection_predictor standalone (before the user has to do the padding removal on it's on which is not correct)
• update API now the detection results are the same with ocr_predictor and detection_predictor

Any feedback is welcome 🤗

CC @ArsalanYounus007 @yumikim381

Note: again API will fail until merged ^^

ref.: #1619 #1622

[felixdittrich92]

comparison with ocr_predictor:

"words": [
    {
        "value": "Hello",
        "geometry": [0.7471996155154171, 0.1796875, 0.8272978149561669, 0.20703125],
        "confidence": 1,
        "crop_orientation": {"value": 0, "confidence": None},
    },
    {
        "value": "world!",
        "geometry": [
            0.8176307908857315,
            0.1787109375,
            0.9101580212741838,
            0.2080078125,
        ],
        "confidence": 1,
        "crop_orientation": {"value": 0, "confidence": None},
    },
],

"words": [
    {
        "value": "Hello",
        "geometry": [
            0.7453157305717468,
            0.1800299733877182,
            0.8233299851417542,
            0.17740298807621002,
            0.8250390291213989,
            0.2027825564146042,
            0.7470247745513916,
            0.20540954172611237,
        ],
        "confidence": 0.99,
        "crop_orientation": {"value": 0, "confidence": 1},
    },
    {
        "value": "world!",
        "geometry": [
            0.8157618045806885,
            0.18002046644687653,
            0.9063061475753784,
            0.17740710079669952,
            0.9078840017318726,
            0.20474515855312347,
            0.8173396587371826,
            0.20735852420330048,
        ],
        "confidence": 1,
        "crop_orientation": {"value": 0, "confidence": 1},
    },
],

[odulcy-mindee]

Why geometry changed so much ?

[felixdittrich92]

@odulcy-mindee this was wrong before because remove padding was only in ocr_predictor (after detection predictor)

Now it's correct the detection results are the same with detection_predictor standalone and ocr_predictor

[felixdittrich92]

ref.: #1619 #1622 :)

[odulcy-mindee]

It's clearer, thank you

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 96.38%. Comparing base (9c92f5c) to head (c1d6a36).

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1627      +/-   ##
==========================================
+ Coverage   96.25%   96.38%   +0.13%     
==========================================
  Files         163      164       +1     
  Lines        7707     7716       +9     
==========================================
+ Hits         7418     7437      +19     
+ Misses        289      279      -10     

Flag	Coverage Δ
unittests	96.38% <100.00%> (+0.13%)	⬆️

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.

[felixT2K]

Included mypy fixes for new version