flatironinstitute/PointCloud_Regression

Point Cloud regression with new algebraic representation on ModelNet dataset (ICCV 2023)

PointCloud_Regression

Point Cloud regression with new algebraical representation on ModelNet40 datasets (ICCV 2023)

Our representation illustrates how quaternion space in 2D must be covered by multiple alternative solutions depending on the actual 2D cosine and sine rotation parameters.

Related Publications

Algebraically rigorous quaternion framework for the neural network pose estimation problem
Chen Lin, Andrew Hanson, and Sonya Hanson
ICCV 2023 [paper]

Get started

Use our architecture on the Jupyter Notebook

start by creating a virtual environment, i.e.

conda create -n env_name

in your terminal. Then activate the environment via:

conda activate env_name

In the environment, first, install our package:

pip install -e .

the dependencies should automatically be installed via setup.py; Alternatively, you could create a docker image.

Install the ipykernel via conda install ipykernel`.

Create a new kernel for your environment by running the command python -m ipykernel install --user --name=env_name.

Start Jupyter Notebook using the command Jupyter notebook. In the Jupyter Notebook interface, you should now see "env_name" as an option in the kernel dropdown menu.

Finally, you could import any module in our code base as a regular package:

from regression.model import PointNet
import regression.metric as M
import regression.adj_util as A
from regression.dataset import SimulatedDataset, ModelNetDataset

Finally, one could run the test via pytest /your/repository/path/PointCloud_Regression/tests/

The test pipeline tested the convention of quaternion, where our homemade code uses Hamitonian convention, but the scipy.spatial.transform uses JPL convention