Skip to content
Code for Leveraging Shape Completion for 3D Siamese Tracking
Branch: master
Clone or download
Latest commit c24280f Apr 15, 2019
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
.vscode clean commits Mar 28, 2019
code clean commits Mar 28, 2019
models clean commits Mar 28, 2019
.gitignore clean commits Mar 28, 2019
Paper.pdf add paper and poster Apr 15, 2019
Poster.pdf add paper and poster Apr 15, 2019
README.md minor fix Apr 15, 2019
ShapeCompletion3DTracking.code-workspace clean commits Mar 28, 2019
environment.yml clean commits Mar 28, 2019

README.md

Leveraging Shape Completion for 3D Siamese Tracking

Supplementary Code for the CVPR'19 paper entitled Leveraging Shape Completion for 3D Siamese Tracking

Supplementary Video

Citation

@InProceedings{Giancola_2018_CVPR,
author = {Giancola, Silvio and Zarzar, Jesus and Ghanem, Bernard},
title = {Leverage Shape Completion for 3D Siamese Tracking},
booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2019}
}

Usage

Download KITTI Tracking dataset

Download the dataset from KITTI Tracking.

You will need to download the data for velodyne, calib and label_02.

Place the 3 folders in the same parent folder as following:

[Parent Folder]
--> [calib]
    --> {0000-0020}.txt
--> [label_02]
    --> {0000-0020}.txt
--> [velodyne]
    --> [0000-0020] folders with velodynes .bin files

Create Environment

conda create -y -n ShapeCompletion3DTracking python tqdm numpy pandas shapely matplotlib pomegranate
source activate ShapeCompletion3DTracking
conda install -y pytorch=0.4.1 cuda90 -c pytorch
pip install pyquaternion

Train a model

python main.py --train_model --model_name=<Name of your model> --dataset_path=<Path to KITTI Tracking folder>

Test a model

python main.py --test_model --model_name=<Name of your model> --dataset_path=<Path to KITTI Tracking folder>

Options

Run python main.py --help for a detailled description of the parameters.

OPT:
    --model_name=<Name of your model>
    --dataset_path=<Path to KITTI Tracking>
    --lambda_completion=1e-6: balance between tracking and completion loss
    --bneck_size=128: lenght of the latent vector
    --GPU=1: enforce the use of GPU 1 
    --tiny: use a tiny set of KITTI Tracking
You can’t perform that action at this time.