GPU Accelerated Non-rigid ICP for surface registration

Last update: Jan 4, 2023

Related tags

Overview

GPU Accelerated Non-rigid ICP for surface registration

Introduction

Preivous Non-rigid ICP algorithm is usually implemented on CPU, and needs to solve sparse least square problem, which is time consuming. In this repo, we implement a pytorch version NICP algorithm based on paper Amberg et al. Detailedly, we leverage the AMSGrad to optimize the linear regresssion, and then found nearest points iteratively. Additionally, we smooth the calculated mesh with laplacian smoothness term. With laplacian smoothness term, the wireframe is also more neat.

Quick Start

install

We use python3.8 and cuda10.2 for implementation. The code is tested on Ubuntu 20.04.

The pytorch3d cannot be installed directly from pip install pytorch3d, for the installation of pytorch3d, see pytorch3d.
For other packages, run

pip install -r requirements.txt

For the template face model, currently we use a processed version of BFM face model from 3DMMfitting-pytorch, download the BFM09_model_info.mat from 3DMMfitting-pytorch and put it into the ./BFM folder.
For demo, run

python demo_nicp.py

we show demo for NICP mesh2mesh and NICP mesh2pointcloud. We have two param sets for registration:

milestones = set([50, 80, 100, 110, 120, 130, 140])
stiffness_weights = np.array([50, 20, 5, 2, 0.8, 0.5, 0.35, 0.2])
landmark_weights = np.array([5, 2, 0.5, 0, 0, 0, 0, 0])

This param set is used for registration on fine grained mesh

milestones = set([50, 100])
stiffness_weights = np.array([50, 20, 5])
landmark_weights = np.array([50, 20, 5])

This param set is used for registration on noisy point clouds

Templated Model

You can also use your own templated face model with manually specified landmarks.

Todo

Currently we write some batchwise functions, but batchwise NICP is not supported now. We will support batch NICP in further releases.

High performance Cross-platform Inference-engine, you could run Anakin on x86-cpu,arm, nv-gpu, amd-gpu,bitmain and cambricon devices.

Anakin2.0 Welcome to the Anakin GitHub. Anakin is a cross-platform, high-performance inference engine, which is originally developed by Baidu engineer

514 Dec 28, 2022

GrabGpu_py: a scripts for grab gpu when gpu is free

GrabGpu_py a scripts for grab gpu when gpu is free. WaitCondition: gpu_memory

3 Jun 18, 2022

A Robust Non-IoU Alternative to Non-Maxima Suppression in Object Detection

Confluence: A Robust Non-IoU Alternative to Non-Maxima Suppression in Object Detection 1. 介绍用以替代 NMS，在所有 bbox 中挑选出最优的集合。 NMS 仅考虑了 bbox 的得分，然后根据 IOU 来

44 Sep 15, 2022

A non-linear, non-parametric Machine Learning method capable of modeling complex datasets

Fast Symbolic Regression Symbolic Regression is a non-linear, non-parametric Machine Learning method capable of modeling complex data sets. fastsr aim

3 Jun 22, 2022

Code for "Learning to Segment Rigid Motions from Two Frames".

rigidmask Code for "Learning to Segment Rigid Motions from Two Frames". ** This is a partial release with inference and evaluation code.

157 Nov 21, 2022

Weakly Supervised Learning of Rigid 3D Scene Flow

Weakly Supervised Learning of Rigid 3D Scene Flow This repository provides code and data to train and evaluate a weakly supervised method for rigid 3D

124 Dec 27, 2022

Official PyTorch implementation of CAPTRA: CAtegory-level Pose Tracking for Rigid and Articulated Objects from Point Clouds

CAPTRA: CAtegory-level Pose Tracking for Rigid and Articulated Objects from Point Clouds Introduction This is the official PyTorch implementation of o

96 Dec 7, 2022

Brax is a differentiable physics engine that simulates environments made up of rigid bodies, joints, and actuators

Brax is a differentiable physics engine that simulates environments made up of rigid bodies, joints, and actuators. It's also a suite of learning algorithms to train agents to operate in these environments (PPO, SAC, evolutionary strategy, and direct trajectory optimization are implemented).

1.5k Jan 2, 2023

Code for ICCV 2021 paper: ARAPReg: An As-Rigid-As Possible Regularization Loss for Learning Deformable Shape Generators..

ARAPReg Code for ICCV 2021 paper: ARAPReg: An As-Rigid-As Possible Regularization Loss for Learning Deformable Shape Generators.. Installation The cod

132 Nov 28, 2022

Comments

Lack of file “BFM09_model_info.mat”

Traceback (most recent call last): File "demo_nicp.py", line 28, in bfm_meshes, bfm_lm_index = load_bfm_model(torch.device('cuda:0')) File "/data/pytorch-nicp/bfm_model.py", line 15, in load_bfm_model bfm_meta_data = loadmat('BFM/BFM09_model_info.mat') File "/root/anaconda3/envs/pytorch3d/lib/python3.8/site-packages/scipy/io/matlab/mio.py", line 224, in loadmat with _open_file_context(file_name, appendmat) as f: File "/root/anaconda3/envs/pytorch3d/lib/python3.8/contextlib.py", line 113, in enter return next(self.gen) File "/root/anaconda3/envs/pytorch3d/lib/python3.8/site-packages/scipy/io/matlab/mio.py", line 17, in _open_file_context f, opened = _open_file(file_like, appendmat, mode) File "/root/anaconda3/envs/pytorch3d/lib/python3.8/site-packages/scipy/io/matlab/mio.py", line 45, in _open_file return open(file_like, mode), True FileNotFoundError: [Errno 2] No such file or directory: 'BFM/BFM09_model_info.mat'

In 3DMMfitting-pytorch， there are only these files: BFM_exp_idx.mat BFM_front_idx.mat facemodel_info.mat README.md select_vertex_id.mat similarity_Lm3D_all.mat std_exp.txt

opened by 675492062 2
What is the expected time needed for running demo_nicp.py?

Hello,

On my computer it seems quite slow to run demo_nicp.py. At least it took more than 1 minutes to get final.obj. Is it correct?

I ranAMM_NRR for non-rigit ICP registration with two 7000 vertices meshes. It needs ca 1 second with CPU on my computer. With GPU, it might be possible to do the same work in less than 100 ms?

Thank you!

opened by 1939938853 0
Hi, with landmarks: `landmarks = torch.from_numpy(np.array(landmarks)).to(device).long()`, maybe you can reshape landmarks from torch.Size([1, 1, 68, 2]) to torch.Size([1, 68, 2])
Hi, with landmarks: landmarks = torch.from_numpy(np.array(landmarks)).to(device).long(), maybe you can reshape landmarks from torch.Size([1, 1, 68, 2]) to torch.Size([1, 68, 2])

Originally posted by @wuhaozhe in https://github.com/wuhaozhe/pytorch-nicp/issues/3#issuecomment-971453681 hi！I got output as torch.Size([1, 68, 512, 3]) torch.Size([1, 68, 2]) torch.Size([1, 512, 512, 3]) I think the shape of following tensors are right, but I meet the same problem. lm_vertex = torch.gather(lm_vertex, 2, column_index) RuntimeError: CUDA error: device-side assert triggered

landmarks = torch.from_numpy(np.array(landmarks)).to(device).long() row_index = landmarks[:, :, 1].view(landmarks.shape[0], -1) column_index = landmarks[:, :, 0].view(landmarks.shape[0], -1) row_index = row_index.unsqueeze(2).unsqueeze(3).expand(landmarks.shape[0], landmarks.shape[1], shape_img.shape[2], shape_img.shape[3]) column_index = column_index.unsqueeze(1).unsqueeze(3).expand(landmarks.shape[0], landmarks.shape[1], landmarks.shape[1], shape_img.shape[3]) print(row_index.shape, landmarks.shape, shape_img.shape)
opened by alicedingyueming 1
RuntimeError

Traceback (most recent call last): File "demo_nicp.py", line 27, in target_lm_index, lm_mask = get_mesh_landmark(norm_meshes, dummy_render) File "/data/pytorch-nicp/landmark.py", line 37, in get_mesh_landmark row_index = row_index.unsqueeze(2).unsqueeze(3).expand(landmarks.shape[0], landmarks.shape[1], shape_img.shape[2], shape_img.shape[3]) RuntimeError: The expanded size of the tensor (1) must match the existing size (2) at non-singleton dimension 1. Target sizes: [1, 1, 512, 3]. Tensor sizes: [1, 2, 1, 1]

I have already configure the environment，but it seems have some problems in the code.What can I do to solve this problem.

opened by 675492062 8

Releases(v0.1)

v0.1(Oct 29, 2021)

We implement mesh2pointcloud and mesh2mesh NICP. Currently, the batch-wise NICP is not supported.
Source code(tar.gz)
Source code(zip)

GPU Accelerated Non-rigid ICP for surface registration

Related tags

Overview

GPU Accelerated Non-rigid ICP for surface registration

Introduction

Quick Start

install

Templated Model

Todo

You might also like...

High performance Cross-platform Inference-engine, you could run Anakin on x86-cpu,arm, nv-gpu, amd-gpu,bitmain and cambricon devices.

GrabGpu_py: a scripts for grab gpu when gpu is free

A Robust Non-IoU Alternative to Non-Maxima Suppression in Object Detection

A non-linear, non-parametric Machine Learning method capable of modeling complex datasets

Code for "Learning to Segment Rigid Motions from Two Frames".

Weakly Supervised Learning of Rigid 3D Scene Flow

Official PyTorch implementation of CAPTRA: CAtegory-level Pose Tracking for Rigid and Articulated Objects from Point Clouds

Brax is a differentiable physics engine that simulates environments made up of rigid bodies, joints, and actuators

Code for ICCV 2021 paper: ARAPReg: An As-Rigid-As Possible Regularization Loss for Learning Deformable Shape Generators..

Comments

Lack of file “BFM09_model_info.mat”

What is the expected time needed for running demo_nicp.py?

Hi, with landmarks: `landmarks = torch.from_numpy(np.array(landmarks)).to(device).long()`, maybe you can reshape landmarks from torch.Size([1, 1, 68, 2]) to torch.Size([1, 68, 2])

RuntimeError

Releases(v0.1)

v0.1(Oct 29, 2021)

Owner

Haozhe Wu

Open source repository for the code accompanying the paper 'Non-Rigid Neural Radiance Fields Reconstruction and Novel View Synthesis of a Deforming Scene from Monocular Video'.

The code for the CVPR 2021 paper Neural Deformation Graphs, a novel approach for globally-consistent deformation tracking and 3D reconstruction of non-rigid objects.

Official code release for ICCV 2021 paper SNARF: Differentiable Forward Skinning for Animating Non-rigid Neural Implicit Shapes.

the code of the paper: Recurrent Multi-view Alignment Network for Unsupervised Surface Registration (CVPR 2021)

Registration Loss Learning for Deep Probabilistic Point Set Registration

GPU-Accelerated Deep Learning Library in Python

GPU-accelerated PyTorch implementation of Zero-shot User Intent Detection via Capsule Neural Networks

NVIDIA Merlin is an open source library providing end-to-end GPU-accelerated recommender systems, from feature engineering and preprocessing to training deep learning models and running inference in production.

(Py)TOD: Tensor-based Outlier Detection, A General GPU-Accelerated Framework

Multiple types of NN model optimization environments. It is possible to directly access the host PC GUI and the camera to verify the operation. Intel iHD GPU (iGPU) support. NVIDIA GPU (dGPU) support.