GitLab

* Code for Paper "Bridging the Gap Between Anchor-based and Anchor-free Detection via Adaptive Training Sample Selection"

* fix format

* changed the code according to the Changes requested

* update benchmark

* minor refactoring

* minor fix

* update model_zoo and support models
Co-authored-by: Cao Yuhang <yhcao6@gmail.com>

* init logger before constructing Runner

* use mmdet logger for loading checkpoints

* bug fix for abstract methods

* Allow for images to contain zero true detections

* Allow for empty assignment in PointAssigner

* Allow ApproxMaxIouAssigner to return an empty result

* Fix CascadeRNN forward when entire batch has no truth

* Correctly assign boxes to background when there is no truth

* Fix assignment tests

* Make flatten robust

* Fix bbox loss with empty pred/truth

* Fix logic error in BBoxHead.loss

* Add tests for empty truth cases

* tests faster rcnn empty forward

* Skip roipool forward tests if torchvision is not installed

* Add tests for bbox/anchor heads

* Consolidate test_forward and test_forward2

* Fix assign_results.labels = None when gt_labels is given; Add test for this case

* Fix OHEM Sampler with zero truth

* remove xdev

* resolve 3 reviews

* Fix flake8

* refactoring

* fix yaml format

* add filter flag

* minor fix

* delete redundant code in load anno

* fix flake8 errors

* quick ...

* async inference support

* implemented concurrent decorator

* fixes for Python versions < 3.7

* async methods depend on python version

* revert changes in forward method

* async_test -> async_simple_test, debug logging is done via logger.debug

* add async test

* add asynctest to test requirements

* async tests are run in Python 3.7

* check CUDA, add docs

* fix device

* run test only if CUDA is available

* fix linting

* custom operators can run on nondefault stream

* set current stream in kernel launch configuration

* example fixes

* add async/sync interface comparison benchmark

* fix linting

* Remove keep all stage code

* remove keep_all_stage in config

* fix scale factor bug

* reformat

* clean code

* reformat for yapf0.29

* fix conflicts of yapf and flake8

* flake 8 is prior than yapf

* Fix simple_test for base and fast_rcnn detector

* Add test of detector forward methods

* Revert forward_test change and add docs indicating expected inputs for forward

* fix linter

* Expand docs

* Fix flake8 errors

* Clarify Collect docstring

* revert to the original version

* add foveabox support

* format

* fix

* foveabox correct

* fix mean and std in config

* add foveabox in model zoo

* fix work_dir, typo, fovea_target and config issues

* add necessary comments

* fix yapf issue in C1 check

* fix yapf issue in C1 check

* rename configs

* update model urls and minor fix

* add tta to HTC and Caccade RCNN

* format file with yapf

* fix import error with isort

* Update htc.py

* Update cascade_rcnn.py

* fix bug

* delete some redundant codes

This reverts commit ea9b20af.

* add reppoints head, reppoints detector and the config files.

* add reppoints generator, target and the center-based assigner.

* add readme for RepPoints

* define data pipelines

* update two config files

* minor fix for config files

* allow img_scale to be optional and update config

* add some docstrings

* add extra aug to transform

* bug fix for mask resizing

* fix cropping

* add faster rcnn example

* fix imports

* fix robustness testing

* add img_norm_cfg to img_meta

* fix the inference api with the new data pipeline

* fix proposal loading

* delete args of DefaultFormatBundle

* add more configs

* update configs

* bug fix

* add a brief doc

* update gt_labels in RandomCrop

* fix key error for new apis

* bug fix for masks of crowd bboxes

* add argument data_root

* minor fix

* update new hrnet configs

* update docs

* rename MultiscaleFlipAug to MultiScaleFlipAug

* add __repr__ for all transforms

* move DATA_PIPELINE.md to docs/

* fix image url

* add flops counter

* minor fix

* add forward_dummy() for most detectors

* add documentation for some tools

* fix the imports in __init__.py

* update contribution guide

* add isort config

* use isort to sort imports

* add isort to travis

* add major components for training double head

double head only with two losses and no attention

remove double_head detector for now, merge upchannel to double_head bbox head

change the stype using yapl

remove uncessary comment#

to pass check

to pass v2

to pass v3

line too long and style again

reuse bottlenet

* refactoring

* bug fix

* bug fix for weight initialization

* add reg roi scale factor and modify loss weights

* rescale the roi after mapping to fpn levels

* format the codebase with yapf

* add yapf to travis

* add fp16 support

* fpn donot need bn normalize

* refactor wrapped bn

* fix bug of retinanet

* add fp16 ssd300 voc, cascade r50, cascade mask r50

* fix bug in cascade rcnn testing

* add support to fix bn training

* add fix bn cfg

* delete fixbn cfg, mv fixbn fp16 to a new branch

* fix cascade mask fp16 bug in test

* fix bug in cascade mask rcnn fp16 test

* add more fp16 cfgs

* add fp16 fast-r50 and faster-dconv-r50

* add fp16 test, minor fix

* clean code

* fix config work_dir name

* add patch func, refactor code

* fix format

* clean code

* move convert rois to single_level_extractor

* fix bug for cascade mask, the seg mask is ndarray

* refactor code, add two decorator force_fp32 and auto_fp16

* add fp16_enable attribute

* add more comment, fix format and test assertion

* fix pep8 format error

* format commont and api

* rename distribute to distributed, fix dict copy

* rename function name

* move function, add comment

* remove unused parameter

* mv decorators into decorators.py, hook related functions to hook

* add auto_fp16 to forward of semantic head

* add auto_fp16 to all heads and fpn

* add docstrings and minor bug fix

* simple refactoring

* bug fix for patching forward method

* roi extractor in fp32 mode

* fix flake8 error

* fix ci error

* add fp16 support to ga head

* remove parallel test assert

* minor fix

* add comment in build_optimizer

* fix typo in comment

* fix typo enable --> enabling

* udpate README

* Grid R-CNN

* add grid_rcnn_res50fpn2x config

* add assertion that grid_head should exist

* fix bugs and remove SharedFCBBoxHeadGrid

* remove the property with_grid

* format fixes for grad_head and add config dir

* move random_jitter to grid_head and some refactoring

* simplify the calculation of num_edges

* refactoring

* refactoring

* rename config files and add x101 config

* bug fix for inference

* remove random_jitter_single

* add readme of grid rcnn

* add bibtex of grid rcnn plus

* update work_dir

* add ms-rcnn

* fix config

* fix bug in test

* fix bug in test, det_labels should plus 1

* add area ratio

* speed up area ratio calculation

* delete caffe style config

* refactor code, bbox and segm eval using different score

* fix bug in distevalhook

* update README and config

* update x101-64x4d

* refactor code

* update upstream

* format

* update r50 and r101 2x performance

* update ms-rcnn-x101-64x4d-2x performance

* move simple_mask_test, add docstring and comment

* delete TODO in mask iou target

* add detail comment to mask iou target

* modify comment of mask iou target

* move mask iou target to mask iou head

* merge mask_iou_target and get_target

* add MSELoss for MaskIoUHead and rename mask_iou to fc_mask_iou

* refactoring

* fix bug in test

* fix bug when cropping mask inside proposal

* update readme

* update bibtex

* remove space in bibtex

* fixed a bug which causes merged_proposals are only from a certain scale of proposals in aug_test_rpn

* add a comment and modify _img_metas to aug_img_metas

* Reformat use yapf 0.27.0

* Add yapf style file

* update the high-level inference api

* save classes in meta data and use it for visualization

* add fcos

* use P5 instead of C5

* add relu before extra convs in FPN

* add singleclass_nms, use caffe2 lr

* fix log interval

* use caffe2init and relu in extra layers

* fix scale layer, use p5 instead of c5

* fix focs target

* refactor code

* delete useless file

* clean

* refactor code

* change num_classes to cls_out_channels

* fix bug of in get_bboxes

* fix bug in test

* add r101 2x cfg

* ms use value mode, add x101-64x4d cfg

* add more comment and rename some variable

* rename centers to points, modify doc string of distance2bbox

* add fcos detector, replace frozen with requires_grad

* add README.md

* add r101-1x performance, rename cfg, add detector FCOS

* update fcos r50 2x performance, remove fpn caffe2 initialize

* fix flake8 error

* rename cfg

* fix grammar error of some comments

* minor fix comment

* change work_dir to be consistent with config name

* add FCOS support in README

* use different configs for proposal train/test

* update configs for dcn

* update inf speed

* update inf speed in dcn, gn, htc

* update comparision

* keep backward compatibility

* add HybridTaskCascade

* add configs for other backbones

* add keep_ratio argument for segmap transform

* add readme for HTC

* fix linting errors

* split assign and sampling as in Cascade R-CNN

* remove unused imports

* add a large model

* update htc

* add two stage w/o neck and w/ upperneck

* add rpn r50 c4

* update c4 configs

* fix

* config update

* update config

* minor update

* mask rcnn support c4 train and test

* lr fix

* cascade support upper_neck

* add cascade c4 config

* update config

* update

* update res_layer to new interface

* refactoring

* c4 configs update

* refactoring

* update rpn_c4 config

* rename upper_neck as shared_head

* update

* update configs

* update

* update c4 configs

* update according to commits

* update

* fix OHEM with cascade_rcnn

* fix OHEM with cascade_rcnn

* delete space

* delete white space

* delete unused lib

* Delete cascade_rcnn_ohem_r101_fpn_1x.py

* fix unreasonable code

* fix Single quote

* fix code style

* fix code style

* fix file permission