search

OpenGVLab / InternImage

[CVPR 2023 Highlight] InternImage: Exploring Large-Scale Vision Foundation Models with Deformable Convolutions
https://arxiv.org/abs/2211.05778
MIT License
2.47k stars 231 forks source link

DCNv3_pytorch代码中可变形卷积的卷积过程 #264

Open lp-094 opened 9 months ago

lp-094 commented 9 months ago

请问在DCNv3_pytorch代码中： output = (samplinginput mask).sum(-1).view(N_, groupgroup_channels, H_outW_out)这一步对应公式里边m_gk x_g, 而x = self.output_proj(x)这一步对应于w_g (m_gk x_g)这一步吗？