Mask RCNN and more

[0x00b1]

In additional to the numerous fixes (and tests), a number of important features are also included:

• a Data Science Bowl (DSB) 2018 dataset
• refactored loss layers
• an entirely rewritten ObjectDetectionGenerator
• a Mask RCNN implementation

[codecov-io]

Codecov Report

Merging #164 into master will increase coverage by 6.8%. The diff coverage is 71.13%.

@@            Coverage Diff            @@
##           master     #164     +/-   ##
=========================================
+ Coverage   69.06%   75.87%   +6.8%     
=========================================
  Files          27       28      +1     
  Lines        1028     1057     +29     
=========================================
+ Hits          710      802     +92     
+ Misses        318      255     -63

Impacted Files	Coverage Δ
keras_rcnn/layers/_object_detection.py	21.56% <0%> (ø)	:arrow_up:
keras_rcnn/layers/__init__.py	100% <100%> (ø)	:arrow_up:
...s_rcnn/layers/object_detection/_object_proposal.py	91.66% <100%> (ø)	:arrow_up:
keras_rcnn/datasets/dsb2018.py	100% <100%> (ø)
keras_rcnn/layers/losses/__init__.py	100% <100%> (ø)	:arrow_up:
...s_rcnn/layers/object_detection/_proposal_target.py	100% <100%> (ø)	:arrow_up:
keras_rcnn/layers/_pooling.py	94.87% <100%> (ø)	:arrow_up:
keras_rcnn/models/_rcnn.py	27.5% <3.7%> (+5.93%)	:arrow_up:
...ras_rcnn/layers/object_detection/_anchor_target.py	75.31% <4.16%> (-2.32%)	:arrow_down:
keras_rcnn/preprocessing/_object_detection.py	86.86% <85.48%> (+41.86%)	:arrow_up:
... 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 96b8af6...a23cb2a. Read the comment docs.

[theone4ever]

It seems the bounding boxes are rotated when using ObjectDetectionGenerator with dataset DSB2018. So in _object_detection.py line 189, instead of target_bounding_box = [ minimum_r, minimum_c, maximum_r, maximum_c ], it should be target_bounding_box = [ minimum_c, minimum_r, maximum_c, maximum_r ]

[0x00b1]