detection_fpn no norm in last layer

detection_fpn no norm in last layer

pytorch update to >1.x, python to 3.7.5. custom exts to c++ and cuda 10.1

intermed save for sharing

fix: mrcnn gt masks roi align

fix: mrcnn gt masks rescale before roi align

fix: roi align in gt masks loss mining, map boxes to unnormalized scale

pre gt masks change

pre pytorch update from 0.4.x to 1.x

smooth in dice loss, no double logging consec folds

file logger separate files per fold

file logger separate files per fold

file logger separate files per fold

smooth also in dice loss

python3.6 in readme

use batchgenerators 19.3

Merge branch 'torch1x'

exec logger fix

issue #91

re-added assets

realigning file tracking

fix retina unet norm in final conv, toy_exp class_spec_seg_flag

consecutive fold-iterations fix, --dev

premerge

readme and make exp parent dirs

readme and make exp parent dirs

public branch update to torch 1.3, python 3.7

torch 1.3.1 and python 3.7 finalization

torch 1.3.1 and python 3.7 finalization

torch 1.3.1 and python 3.7 finalization

re-added assets

max patients in defconfigs

requirements fix

rm test error

readme, unittests

pre python reinstall

reqs

fixes, nms behaviour like pytorch (no +1), unittests

unresolved no program exit if no sysmetrics logging

hangup present

tracking exit hangup

autom exts install

tracking exit hangup

tracking exit hangup

copyright in readme

retina net anchor match fix

seg default bugs, searching

tracking exit hangup related matplotlib

undefault sysmetrics logging

shorter --dev cycle

readme fix

readme fix

re-added assets

re-added assets

rm code analysis

setup cleanup

setup cleanup

readme fix

exts setup fix

merge master into public

exts setup fix

save

added .gitignore

install batchgenerators via pip

readme typo fix

egg info cleaning

cf change

make parent experiment dirs

tboard in reqs

automatic cust exts install

automatic install of custom exts

requirements clean up

automatic install of custom exts

implemented new RoIAl and NMS into model and utils interfaces, removed tboardX

nms 3d ext fix (blockboxes threads), need to figure extreme cases in roialign

ra3d forw tested vs. numpy implement, nms swapped in models for new version…

RA3D: fix in unittest (torch requires weird format for input rois )

RA3D: backward written in CUDA, needs testing and finalization

RA3D: forward finished, needs testing, backward still pending

roi align 2d compiles as extension, 3D in progress

nms-extension interface more inline w/ pytorch

tested nms, cuda & own numpy fine, slight inconsistency in torch.ops.nms on cpu

nms cuda extension finished, now has 3D capability

retina_net docstring fix

2D torch nms as cpp extension compiled from source + analysis file for…

nms rewriting in c++, cuda in the very beginning

retina_net docstring fix

public init commit

Reg RCNN private master initial commit

Reg RCNN private master initial commit