lmb-freiburg / sf2se3

Repository for SF2SE3: Clustering Scene Flow into SE(3)-Motions via Proposal and Selection

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

SF2SE3

SF2SE3: Clustering Scene Flow into SE(3)-Motions via Proposal and Selection @ DAGM German Conference on Pattern Recognition 2022.

Algorithm

To understand the algorithm start here.

Setup

(a) Docker

sudo docker build -t sflow2se3:v1 -f ./scripts/setup/Dockerfile .

(b) Local

bash scripts/setup/setup_sflow2se3.sh
source venv_py38_sflow2se3/bin/activate

Datasets

(a) Custom Stereo : supplement datasets/Custom_Stereo
(b) Custom RGBD : supplement datasets/Custom_RGBD
(c) FlyingThings3D : download (Subset DispNet/FlowNet2.0) and adapt & run
(d) KITTI-2015 : download
(e) TUM RGBD : download

Inference

(a) Docker

sudo docker run --gpus all -v <ext_vol>:<ext_vol> sflow2se3:v1 <configs>

(b) Local

python3.8 eval.py <configs>

Configurations

For setup adjustments look at config-setup e.g. datasets directory and output directory.
For data adjustments look at config-data e.g. camera settings.

(a) Custom Stereo

--config-setup configs/setup/custom.yaml --config-data configs/data/custom_stereo.yaml --config-sflow2se3 configs/sflow2se3/classic.yaml --config-sflow2se3-data-dependent configs/sflow2se3/custom_stereo/classic.yaml

Note: Data is expected to have same data types as FlyingThings3D. Otherwise, adjustements to the data loading is necessary.

(b) Custom RGBD

--config-setup configs/setup/custom.yaml --config-data configs/data/custom_rgbd.yaml --config-sflow2se3 configs/sflow2se3/classic.yaml --config-sflow2se3-data-dependent configs/sflow2se3/custom_rgbd/classic.yaml

Note: Data is expected to have same data types as TUM RGBD. Otherwise, adjustements to the data loading is necessary.

(c) FlyingThings3D

--config-setup configs/setup/custom.yaml --config-data configs/data/flyingthings3d_dispnet.yaml --config-sflow2se3 configs/sflow2se3/classic.yaml --config-sflow2se3-data-dependent configs/sflow2se3/flyingthings3d_dispnet/classic.yaml

(d) KITTI-2015

--config-setup configs/setup/custom.yaml --config-data configs/data/kitti.yaml --config-sflow2se3 configs/sflow2se3/classic.yaml --config-sflow2se3-data-dependent configs/sflow2se3/kitti/classic.yaml

(e) TUM RGBD

--config-setup configs/setup/custom.yaml --config-data configs/data/tum_rgbd_fr3.yaml --config-sflow2se3 configs/sflow2se3/classic.yaml --config-sflow2se3-data-dependent configs/sflow2se3/tum_rgbd_fr3/classic.yaml

alt text alt text

Credits

Optical Flow Neural Network : RAFT
Disparity Neural Network : LEAStereo

About

Repository for SF2SE3: Clustering Scene Flow into SE(3)-Motions via Proposal and Selection


Languages

Language:Python 97.1%Language:Shell 2.9%Language:Dockerfile 0.1%