Recurrent Multi-view Alignment Network for Unsupervised Surface Registration
RMA-Net
This repo is the implementation of the paper: Recurrent Multi-view Alignment Network for Unsupervised Surface Registration (CVPR 2021).
Paper address: https://arxiv.org/abs/2011.12104
Project webpage: https://wanquanf.github.io/RMA-Net.html
Prerequisite Installation
The code has been tested with Python3.8, PyTorch 1.6 and Cuda 10.2:
conda create --name rmanet
conda activate rmanet
conda install pytorch=1.6.0 torchvision=0.7.0 cudatoolkit=10.2 -c pytorch
conda install -c conda-forge igl
Other requirements include: eigen3, Openmesh and MeshlabServer.
Build the cuda extension:
python build_cuda.py
Usage
Pre-trained Models
Download the pre-trained models and put the models in the [YourProjectPath]/pre_trained folder.
Run the registration
To run registration for a single sample, you can run:
python inference.py --weight [pretrained-weight-path] --src [source-obj-path] --tgt [target-obj-path] --iteration [iteration-number] --device_id [gpu-id] --if_nonrigid [1 or 0]
The