Implementation of Newcombe et al. 2015 DynamicFusion paper.
This project is still in active development and does not yet reproduce the results of the paper accurately.
The code is based on this KinectFusion implemenation
Clone dynamicfusion and dependencies.
git clone https://github.com/mihaibujanca/dynamicfusion --recursive
Install NVIDIA drivers.
- Enable NVidia drivers (Search / Additional Drivers) selecting: "Using NVIDIA binary driver - version 375.66 from nvidia-375 (proprietary, tested)" "Using processor microcode firmware for Intel CPUs from intel-microcode (proprietary)"
- Restart pc to complete installation
Alternatively a good tutorial with some common issues covered can be found here.
For fresh installs (this assumes you cloned your project in your home directory!):
chmod +x build.sh
./build.sh
If you are not on a fresh install, check build.sh for building instructions and dependencies.
If you want to build the tests as well, set -DBUILD_TESTS=ON.
To save frames showing the reconstruction progress, pass -DSAVE_RECONSTRUCTION_FRAMES=ON. The frames will be saved in <project_root>/output
To build documentation, go to the project root directory and execute
doxygen -g
doxygen Doxyfile
./download_data
./build/bin/dynamicfusion data/umbrella
Dependencies:
- CUDA 5.0 or higher
- OpenCV 2.4.8 or higher (modules opencv_core, opencv_highgui, opencv_calib3d, opencv_imgproc, opencv_viz). Make sure that WITH_VTK flag is enabled in CMake during OpenCV configuration.
- Boost (libraries system, filesystem and program options. Only used in the demo. Tested with 1.64.0)
- Ceres solver (Tested with version 1.13.0)
Implicit dependency (needed by opencv_viz):
- VTK 5.8.0 or higher
- SuiteSparse, BLAS and LAPACK for ceres Optional dependencies:
- GTest for testing
- Doxygen for documentation
- OpenNI v1.5.4 for getting input straight from a kinect device.
Install NVIDIA drivers and CUDA
- Install LAPACK.
- Install VTK (download and build from source)
- Install OpenCV.
- Install Boost
Optionals:
Download the dataset.
Create a data folder inside the project root directory.
Unzip the archive into data and remove any files that are not .png.
Inside data, create directories color and depth, and move color and depth frames to their corresponding folders.
To use with .oni captures or straight from a kinect device, use ./build/bin/dynamicfusion_kinect <path-to-oni> or ./build/bin/dynamicfusion_kinect <device_id>
Note: currently, the frame rate is too low (10s / frame) to be able to cope with live inputs, so it is advisable that you capture your input first.
@InProceedings{Newcombe_2015_CVPR,
author = {Newcombe, Richard A. and Fox, Dieter and Seitz, Steven M.},
title = {DynamicFusion: Reconstruction and Tracking of Non-Rigid Scenes in Real-Time},
booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2015}
}
The example dataset is taken from the VolumeDeform project.
@inbook{innmann2016volume,
author = "Innmann, Matthias and Zollh{\"o}fer, Michael and Nie{\ss}ner, Matthias and Theobalt, Christian
and Stamminger, Marc",
editor = "Leibe, Bastian and Matas, Jiri and Sebe, Nicu and Welling, Max",
title = "VolumeDeform: Real-Time Volumetric Non-rigid Reconstruction",
bookTitle = "Computer Vision -- ECCV 2016: 14th European Conference, Amsterdam, The Netherlands,
October 11-14, 2016, Proceedings, Part VIII",
year = "2016",
publisher = "Springer International Publishing",
address = "Cham",
pages = "362--379",
isbn = "978-3-319-46484-8",
doi = "10.1007/978-3-319-46484-8_22",
url = "http://dx.doi.org/10.1007/978-3-319-46484-8_22"
}