Makes Pretrained consistent

[oke-aditya]

Suppresses #33

This does a small refactor for lightning trainers to support pretrained models.

Somehow this issue is not fully resolved. But this at least makes pretrained consistent to str.

We should try to avoid boolean for it.

A complete solution is rather hard. Plus lack of pretrained weights would simply keep creating errors here and there.

If community can contribute resnet_fpn weights. It would be great !! We can then think of adding that feature by changing a bit to the backbone API.

[codecov]

Codecov Report

Merging #47 (bfcf280) into master (06dc0ac) will not change coverage.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##           master      #47   +/-   ##
=======================================
  Coverage   78.06%   78.06%           
=======================================
  Files          41       41           
  Lines        1372     1372           
=======================================
  Hits         1071     1071           
  Misses        301      301           

Flag	Coverage Δ
unittests	78.06% <100.00%> (ø)

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

Impacted Files	Coverage Δ
quickvision/models/detection/detr/model_factory.py	100.00% <100.00%> (ø)
.../models/detection/faster_rcnn/lightning_trainer.py	100.00% <100.00%> (ø)
...on/models/detection/retinanet/lightning_trainer.py	100.00% <100.00%> (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 06dc0ac...bfcf280. Read the comment docs.