Skip to content

Deformable Convolution Networks(deformable conv2D TF_implementation)

Notifications You must be signed in to change notification settings

Worromots/custom_ops

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

21 Commits
 
 
 
 
 
 

Repository files navigation

这个库主要实现了可变形卷积V2和可变形池化的CPU和GPU版本

目前deformable_conv2d_ops_new目录下的实现包含了DeformablePSROIAlign和DeformableConv2D的实现,并对cpu和gpu代码的进行了分离,从而使用分离编译,NVCC编译gpu部分,gcc或者clang编译cpu部分,从而提高在cpu部分代码的performance,推荐使用该目录下的代码

目前是在tensorflow2.0-gpu上编译通过,在tensorflow1.14-gpu略做修改也可以编译通过,后续考虑添加一个支持tensorflow1.14的分支。

如果需要自己编译使用,修改CMakeLists中的${TF_INCLUDE_PATH}变量 和最后target_link_libraries(ops PUBLIC ${CUDA_LIBRARIES} /home/admin-seu/miniconda3/envs/py36/lib/python3.6/site-packages/tensorflow_core/libtensorflow_framework.so.2) 中最后手动给定的lib位置,这里给出的是我编译时用的lib位置,对于你自己使用的tensorflow库你需要修改为你目录下的地址, USE_CUDA默认设置为开启,cmake -DUSE_CUDA=OFF可以关闭对gpu部分的编译。

编译结束后使用deformable_conv2d.py进行测试

batch_conv: 使用depth_wise函数来模拟对和batch有关系的卷积操作

CARAFE: https://arxiv.org/pdf/1905.02188.pdf CARAFE: Content-Aware ReAssembly of Features 根据这篇文章提出的算子实现,ICCV2019 oral论文

2020.1.4 update

add c++ operator deformable conv2d_v2, code is build is tensorflow2.0, No support for other tensorflow version

2020.1.13 update

add c++ operator deformable conv2d_v2(Both CPU and GPU), code is build is tensorflow2.0, No support for other tensorflow version

About

Deformable Convolution Networks(deformable conv2D TF_implementation)

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • C++ 61.0%
  • Makefile 27.2%
  • Cuda 5.1%
  • CMake 2.9%
  • C 2.5%
  • Python 1.3%