[uni-uvpt] About build and pretrain with DCNv2

[WJHash]

Could you please kindly provide the versions of CUDA and PyTorch projects using?
Project needs Deformable-Convolution-V2 (https://github.com/chengdazhi/Deformable-Convolution-V2-PyTorch.git) But it works on torch=0.4.1，and my environment is also unable to use the [DCNv2_pytorch_1.0.0 branch].

[WJHash]

There also has no "MSDeformAttn" module in project Deformable-Convolution-V2

[WJHash]

I build it successfully with torch1.10.2，cuda 10.2，but get error
"python pretrain.py configs/pretrain/daformer_swin_base_patch4_window7_512x512_400k_2e-6_synthia.py"，when pretraining. After searching，I think there is still something wrong with env. Install libtorch cant fix it

[WJHash]

There also has no "MSDeformAttn" module in project Deformable-Convolution-V2. And in install.sh it rm MultiScaleDeformableAttention.egg-info. Is the project use https://github.com/FscoreLab/MultiScaleDeformableAttention.git and not https://github.com/chengdazhi/Deformable-Convolution-V2-PyTorch.git?

[Siris-qy]

I’m facing the same issue. Have you managed to resolve it?