DIM-Net: Learning Detail Disentangled Implicit Fields from Single Images Supplementary Material

In the implementation of DIM-Net, we take ResNet18 as our encoder to obtain the global feature and local feature map from the input image. The base decoder is an MLP with the architecture of IMNET [1]. The detail decoder follows the network in [4] to predict the two displacement maps from the local feature map. As for DIM-NetGL, we take DISN [3] as the base decoder with both their global decoder and local decoder.