vanvalenlab / deepcell-tf / 803

86%

push

travis-ci-com

Support training RetinaNet and MaskRCNN models (#153)

* add custom retinanet layers, anchor utils, and losses

* add RetinaNet Model and custom PriorProbability initializer

* build_ext in Dockerfile and ignore from git

* moved model_zoo.py into model_zoo/featurenet.py and removed fixed field size models

* added retinanet models to model_zoo/retinanet.py and add TensorProduct to fix the 3 channel issue.

* add partition_info to Initializer.__call__

* remove images without any bounding boxes on init()

* `clear_borders` flag for calling clear_border on y data

* update keras-preprocessing to use 1.0.6 for setting interpolation_order

* let norm_method be an optional parameter defaulting to "whole_image"

* support tf.keras.applications backbones

* Implement MaskRCNN (#152)

* remove keras-retinanet and keras-maskrcnn from requirements.txt and travis.yml

* add cython to requirements.txt for building compute_overlap

* install each requirement on a different line, and include keras_preprocessing and networkx

* commenting out FilterDetections test_mini_batch test as it fails randomly.

1093 of 1093 new or added lines in 14 files covered. (100.0%)

2732 of 4050 relevant lines covered (67.46%)

3.37 hits per line