GitLab

原因：
np.allclose(a, b, rtol)使用，计算误差是(a-b)的误差在(atol + rtol*b.abs())以内，但用例中b的值有0，误差就是atol了，atol默认是10的-8.
修改：
把atol根据用例改为传入的误差*10-3.

* Remove runner, parallel, engine and device

* fix format

* remove outdated docs

* migrate many functions to mmengine

* remove sync_bn.py

* add pyupgrade

* add options for pyupgrade

* minor refinement

* first commit

* Add copyright pre-commit-hook

* [Fix] fix deform conv by add argument

* [Fix] replace useless func with np.repeat and add necessary comment

* [Fix] reduce batch_size and remove useless lines and modify some var name

* [Fix] change position of comments

* [Fix] add im2col_step in the docstring and add two test cases

* [Fix] fix docstring and add comments for test cases

* [Fix] fix docstring

* [Fix] add note, fix issue link and other details

* [Fix] fix docstring details

* [Fix] fix links in docstring

* [Fix] fix docstring details

* DCN cpu version

* add modulated dcn cpu version

* move deform_conv_shape_check to deform conv utils

* add inline to deform_conv_shape_check

* add tests

* run linter

* add newline at file end

* run pre-commit against modulated deform conv cpp

* update saconv test

* run clang-format

* remove cuda device inline

* refactor dcn cuda/cpu functions

* remove DCN util

* remove DCN util hpp from all included files

* Addressing PR comment by refactoring modulated-DCN

* fix lint in cpp files

* improve digit_version & use it for version_checking

* more testing for digit_version

* setuptools >= 50 is needed

* fix CI

* add debuging log

* >= to ==

* fix lint

* remove

* add failure case

* replace

* fix

* consider TORCH_VERSION == 'parrots'

* add unittest

* digit_version do not deal with the case if 'parrots' in version name.

* fix fp16 bug on DCNv2

* support fp16 on DCN/DCNv2 when pytorch >= '1.6.0'

* add comment

* Modified the comments

* Unified the usages of '.to()' and '.type_as()'

* [Bug] Fix DeformConv2d bias error and add tests

* fix repr

* revise tests

* lint

* 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>