Deep Cloeset Scene Flow

Introduction to model coming soon!

Prerequisites

Install the dependencies into your virtual environment. To do so:

pip install -r dependencies/requirements.txt

If using conda, the evironment has been exported. To create environment from .yml:

conda env create -f dependencies/environment.yml

Make sure you install cudatoolkit. The code is highly dependent on CUDA!

Additional requirements for visualization:

pip install numba
pip install cffi
sudo apt-get -y install python-vtk
pip install mayavi
sudo apt-get install python3-pyqt5
pip install PyQt5

Training Registration

DCP-v1

python main.py --exp_name=dcp_v1 --model=dcp --emb_nn=dgcnn --pointer=identity --head=svd

DCP-v2

python main.py --exp_name=dcp_v2 --model=dcp --emb_nn=dgcnn --pointer=transformer --head=svd

Testing Registration

DCP-v1

python main.py --exp_name=dcp_v1 --model=dcp --emb_nn=dgcnn --pointer=identity --head=svd --eval

or

python main.py --exp_name=dcp_v1 --model=dcp --emb_nn=dgcnn --pointer=identity --head=svd --eval --model_path=xx/yy

DCP-v2

python main.py --exp_name=dcp_v2 --model=dcp --emb_nn=dgcnn --pointer=transformer --head=svd --eval

or

python main.py --exp_name=dcp_v2 --model=dcp --emb_nn=dgcnn --pointer=transformer --head=svd --eval --model_path=xx/yy

where xx/yy is the pretrained model

Training Scene Flow

./run_flow_training.sh

Testing Scene Flow

./run_flow_testing.sh

Scene Flow Result