Add fasterrcnn example

[oke-aditya]

Please check the tutorial.

If more details about Faster RCNN and its hyperparameters is required, please tell me I will add it.

Also if some theory about this model is required. Will add that as well.

[review-notebook-app]

Check out this pull request on 

Review Jupyter notebook visual diffs & provide feedback on notebooks.

---

Powered by ReviewNB

[codecov-commenter]

Codecov Report

Merging #80 into master will decrease coverage by 4.61%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master      #80      +/-   ##
==========================================
- Coverage   56.00%   51.39%   -4.62%     
==========================================
  Files          55       60       +5     
  Lines        1714     1827     +113     
==========================================
- Hits          960      939      -21     
- Misses        754      888     +134     

Flag	Coverage Δ
#unittests	51.39% <ø> (-4.62%)	⬇️

Impacted Files	Coverage Δ
mantisshrimp/models/detr_demo.py	20.68% <0.00%> (-79.32%)	⬇️
mantisshrimp/models/mantis_detr.py	40.00% <0.00%> (-56.67%)	⬇️
...antisshrimp/models/mantis_rcnn/mantis_mask_rcnn.py	58.62% <0.00%> (-33.38%)	⬇️
mantisshrimp/utils/torch_utils.py	60.00% <0.00%> (-10.00%)	⬇️
mantisshrimp/data/convert_records_to_coco_style.py	85.45% <0.00%> (-8.55%)	⬇️
...tisshrimp/models/mantis_rcnn/mantis_faster_rcnn.py	96.29% <0.00%> (-0.68%)	⬇️
mantisshrimp/imports.py	100.00% <0.00%> (ø)
mantisshrimp/hub/pennfundan/pennfundan_parser.py	0.00% <0.00%> (ø)
mantisshrimp/utils/data_dir.py	87.50% <0.00%> (ø)
mantisshrimp/hub/pennfundan/__init__.py	0.00% <0.00%> (ø)
... and 4 more

---

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 66ebefa...c7d4e36. Read the comment docs.

[lgvaz]

Thanks for the PR!

You should use ReviewNB for checking the code suggestions, it should show nice in notebook changes.

[lgvaz]

There is a lot of overlap between this tutorial and this one.

I do like the part while you over the different models we can use thought.

I have two suggestions:

1

Merge the two tutorials into one. But keep in mind that this is probably going to be the first tutorial beginners see, so try not to overwhelm them with information

2

Transform this into a advanced/extended tutorial, in this case you can even tell how stuff is happening in the inside, going really into detail, don't spare words.

---

I personally like more suggestion 2, which I think is going to be very complementary to the tutorials we already have and will facilitate a lot the lives of people wanting to get involved with the library.

But it's also a lot more work... well, your call at the end 😆

[oke-aditya]

Personally Even I felt I should do 2nd thing. We can have 2 tutorials, first one as a QuickStart. And 2nd an in depth guide for this.

[oke-aditya]

The code block does not look good on ReviewNB. But locally it is looking like a code block
like this

num_classes=None,
# transform parameters
min_size=800, max_size=1333,
image_mean=None, image_std=None,
# RPN parameters
rpn_anchor_generator=None, rpn_head=None,
rpn_pre_nms_top_n_train=2000, rpn_pre_nms_top_n_test=1000,
rpn_post_nms_top_n_train=2000, rpn_post_nms_top_n_test=1000,
rpn_nms_thresh=0.7,
rpn_fg_iou_thresh=0.7, rpn_bg_iou_thresh=0.3,
rpn_batch_size_per_image=256, rpn_positive_fraction=0.5,
# Box parameters
box_roi_pool=None, box_head=None, box_predictor=None,
box_score_thresh=0.05, box_nms_thresh=0.5, box_detections_per_img=100,
box_fg_iou_thresh=0.5, box_bg_iou_thresh=0.5,
box_batch_size_per_image=512, box_positive_fraction=0.25,
bbox_reg_weights=None

[lgvaz]

@ai-fast-track can we get your input?

I feel like there is a lot of overlaps in the tutorials right now, between this one, wheat and parser.

Should we avoid overlap or is it a good thing to have multiple tutorials with different levels of details?

[oke-aditya]

Helps issue #39

[oke-aditya]

@ai-fast-track can you please have a look ?

[ai-fast-track]

@lgvaz @oke-aditya, I think we should add this notebook even if there is some overlap with wheat.ipynb. The latter can be used as quick showcase and highlighted (by exposing its source code) in the README

For this tutorial, I think we should use one of the toy (small) dataset that we discussed about during the last meeting. A dataset that uses the COCO format. That way, this notebook will be different (COCO parser) from the wheat notebook (custom parser), and we can easily and quickly run it in Colab!

[oke-aditya]

@ai-fast-track Suggestions are made. Can you please have another look?