使用RTM3D 代码，CenterTrack代码需要用DCN

1、安装DCNv2

• （1）github上最新版的DCNv2源码在"https://github.com/CharlesShang/DCNv2"，但是该版本源码不支持PyTorch1.7，如果使其支持PyTorch1.7需要做以下修改：
  将DCNv2/src/cuda/dcn_v2_cuda.cu中所有的THCState_getCurrentStream(state)替换成c10::cuda::getCurrentCUDAStream()。

• （2）github上有支持PyTorch1.7的源码，“https://github.com/lbin/DCNv2/tree/pytorch_1.7”，可以直接使用。

2、 Deformable Convolutional Networks V2 with Pytorch 1.0，不支持PyTorch1.7

Build

    ./make.sh         # buildpython test.py    # run examples and gradient check 

An Example

• deformable conv

    from dcn_v2 import DCNinput = torch.randn(2, 64, 128, 128).cuda()# wrap all things (offset and mask) in DCNdcn = DCN(64, 64, kernel_size=(3,3), stride=1, padding=1, deformable_groups=2).cuda()output = dcn(input)print(output.shape)

• deformable roi pooling

    from dcn_v2 import DCNPoolinginput = torch.randn(2, 32, 64, 64).cuda()batch_inds = torch.randint(2, (20, 1)).cuda().float()x = torch.randint(256, (20, 1)).cuda().float()y = torch.randint(256, (20, 1)).cuda().float()w = torch.randint(64, (20, 1)).cuda().float()h = torch.randint(64, (20, 1)).cuda().float()rois = torch.cat((batch_inds, x, y, x + w, y + h), dim=1)# mdformable pooling (V2)# wrap all things (offset and mask) in DCNPoolingdpooling = DCNPooling(spatial_scale=1.0 / 4,pooled_size=7,output_dim=32,no_trans=False,group_size=1,trans_std=0.1).cuda()dout = dpooling(input, rois)

Note

Now the master branch is for pytorch 1.0 (new ATen API), you can switch back to pytorch 0.4 with,

git checkout pytorch_0.4

3、pytorch代码

https://github.com/HiYx/pytorch-deform-conv-v2

4、mmdet 的DCN

可以直接使用mmdet 的DCN。

5、附录

新

旧

6、可参考

https://blog.csdn.net/wangxiao7474/article/details/113605230