GitLab

* [Feature] Add revert_sync_batchnorm

* support mmsyncbn (to be tested)

* Test passed

* Update docstring, rename the test file

* remove test_sync_bn

* add comment

* add mmcv.ops check

* Add a comment

* Add notes and relax test req
Co-authored-by: gaotongxiao <gaotongxiao@gmail.con>

* restrict the warning message

* and an important keyword in warning description

* a more elegant way of condition

* link format code too long

* fix the stupid spelling mistake

* Use issubclass to restrict warning message.

* maybe this version is more elegant.

* conv + bias + norm warning pytest

* 'created' a warning, hahaha

* isort and yapf format revision

* isort and yapf format revision

* flake8 fail issue

* I have to right this way in order to solve the conflicts between yapf and flake8, sigh...

* fixed test bug

* Add ruby windows installer source.

* Simplified the code and remove ruby source from CONTRIBUTING.md

* use _BatchNorm to simplify the code

* bug fix and add instanceNorm case into warning

* change the warning message to make it more clear

* fix unit test

* [Docs] Add header for files

* change to OpenMMLab

* add headers for .cpp, .cu, .h, .cuh

* replace Open-MMLab with OpenMMLab

* empty tensor inference backward continity

* update

* add 3d

* minor changes

* change to modulist

* change to Sequential

* replace dropout with attn_drop and proj_drop in MultiheadAttention

* add operation_name for attn

* add drop path and move all ffn args to ffncfgs

* fix typo

* fix a bug when use default value of ffn_cfgs

* fix ffns

* add deprecate warning

* fix deprecate warning

* change to pop kwargs

* support register FFN of transformer

* support batch first

* fix batch first wapper

* fix forward wapper

* fix typo

* fix lint

* add unitest for transformer

* fix unitest

* fix equal

* use allclose

* fix comments

* fix comments

* change configdict to dict

* move drop to a file

* add comments for drop path

* add noqa 501

* move bnc wapper to MultiheadAttention

* move bnc wapper to MultiheadAttention

* use dep warning

* resolve comments

* add unitest:

* rename residual to identity

* revert runner

* msda residual to identity

* rename inp_identity to identity

* fix name

* fix transformer

* remove key in msda

* remove assert for key
Co-authored-by: HIT-cwh <2892770585@qq.com>
Co-authored-by: bkhuang <congee524@gmail.com>
Co-authored-by: Wenwei Zhang <40779233+ZwwWayne@users.noreply.github.com>

* [Fix] Support names of base classes matching in init_cfg

* revise bool to len

* [Feature] Add truncated normal weight init

* [Feature] Add truncated normal weight init

* [Feature] Add truncated normal weight init

* update docstring

* delete modelA.pth

* modify according to comment

* use kstest to check truncated normal

* delete modelA.pth

* fix test.txt

* fix generalized attention fp16

* fix building without gpu error

* add comment

* Cast tensor at initialization

* [Fix] Config deep copy in initialize_override

* add asserts&comments

* add test

* test org init_cfg

* test override without name

* typo

* add model registry

* fixed infer scoep

* fixed build func

* add docstring

* add md

* support multi level

* clean comments

* add docs

* fixed parent

* add more doc

* add value error, add docstring

* fixed docs

* change to local/global search

* resolve comments

* fixed test

* update some docstring

* update docs (minior)

* update docs

* update docs

* [Refactoring]Approve init_cfg

* revise unit test

* revise according to comments

* add comment for wholemodule

* [Refactoring] Add Caffe2Xavier Initializer

* fix lint

* fix bug of convmodule

* fix bug of convmodule

* fix unitest

* remove assert

* [Feature] support to calculate FLOPs of GN, IN, LN

* Update flops_counter.py

* Update flops_counter.py

* add initializers and BaseModule for unified parameter initialization

* fix circle import

* bug fix

* add is_init flag in BaseModule

* fix docstring

* sort import and fix doc format

* fix bug

* fix docformat and double quote string

* fix import sort

* import sort

* sort import

* revise according to comments

* fix doc format

* revise according to comments

* revise import and fix typo

* polish code

* revise minors

* revice minors

* revise apply function

* revise bias initialization with probability

* add type test for bias_prob

* revise minors

* avoid recursion error when using customized norm layer without _abbr_

* fix unittest bug

* add clamp without unittest

* add clamp-act with unit test

* fix name bug

* use logical and

* fix logical_and

* fix linting

* rename ClampLayer to Clamp

* rename ClampLayer to Clamp
Co-authored-by: nbei <631557085@qq.com>

* enhance hsigmoid

* delete int

* support pytorch 1.7 and update the CI

* fix CI

* fix onnxruntime version in macos

* add version check in wrappers

* fix assersion

* use digital version for version comparison

* fix unit tests

* reformat

* fall back to compare the first two version

* fix unittest

* fix unittest

* fix unit test

* clean unnecessary change

* register deconv in CONV_LAYERS

* use ConvTranspose2d implemented in MMCV

* remove repetitive register_module

* update

* add unittest for deconv

* Mv wrappers into bricks and use wrappers in registry

* resolve import issues

* fix import issues

* set nn op forward to torch 1.6.1

* fix CI bug and add warning

* Fix CI by using patch mock

* mv warnings inside deprecated module's initialization

* update impad

* fix docstring

* add shape for impad

* fix unit test

* remove old version & fix doc

* fix linting

* fix doc

* add linear decay learning rate scheduler

* fix impad

* fix setup.cfg

* fix linting

* add yapf

* add swish

* fix lr_updater

* fix lr_updater.py

* update swish

* add swish

* fix inplace

* fix typo

* update

* add same padding

* fix docstring

* add unittest

* fix register

* change name
Co-authored-by: lixiaojie <lixiaojie@sensetime.com>

* update impad

* fix docstring

* add shape for impad

* fix unit test

* remove old version & fix doc

* fix linting

* fix doc

* add linear decay learning rate scheduler

* fix impad

* fix setup.cfg

* fix linting

* add yapf

* add swish

* fix lr_updater

* fix lr_updater.py

* update swish

* add swish

* fix inplace

* fix typo
Co-authored-by: lixiaojie <lixiaojie@sensetime.com>

* add pairwise function for 'gaussian' and 'concatenation' mode

* rename test function

* decrease the complexity of nonlocal unittest

* fix typo and make unittest more complete

* add unittest when zero_init is False

* minor fix

* pack theta and phi
Co-authored-by: Jiarui XU <xvjiarui0826@gmail.com>

* Add MMSyncBN in registery

* skip mmsyncbn test

* move fuse conv bn to mmcv

* update doc

* update test conv bn

* rename

* fix doc and variable name

* change func name

* Refator plugins: move from ops to cnn

* minior update

* minior update

* fix bug of building ConvModule with HSigmoid using inplace=True and add corresponding unittest

* fix linting

* add ext ops, support parrots

* fix lint

* fix lint

* update op from mmdetection

* support non-pytorch env

* fix import bug

* test not import mmcv.op

* rename mmcv.op to mmcv.ops

* fix compile warning

* 1. fix syncbn warning in pytorch 1.5
2. support only cpu compile
3. add point_sample from mmdet

* fix text bug

* update docstrings

* fix line endings

* minor updates

* remove non_local from ops

* bug fix for nonlocal2d

* rename ops_ext to _ext and _ext to _flow_warp_ext

* update the doc

* try clang-format github action

* fix github action

* add ops to api.rst

* fix cpp format

* fix clang format issues

* remove .clang-format
Co-authored-by: Kai Chen <chenkaidev@gmail.com>

* Add utils to calculate model complexity info

* remove _InstanceNorm in unittest

* add docstring and increase unittest coverage

* fix deconv_flops_counter_hook to accept different data shape

* test when model is not a common instance

* put flops_counter.py and weight_init.py into mmcv/cnn/utils folder

* fix import name

* reformat some docstrings

* update the scripts with latest one and remove redundant codes

* directly represent a model without string and eval()

* reformat code

* add hswish and hsigmoid

* fix linting

* fix linting

* delete useless staticmethod in cnn/bricks/hsigmoid.py

* forward(x) ==> forward(seld, x)

* add unittests for hsigmoid and hswish

* use torch.equal instead of numpy.equal

* add non_local module

* rewrite non local module comments

* perfect docstring and adjust init function

* not to init norm layer

* Correct initialize when there is a norm

* set normal method for both embedded_gaussian and dot_product