Skip to content
This source code provides a reference implementation for ENFT-SfM.
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
3rdparty
ENFT
config
x64
.gitattributes add source code Jul 20, 2018
.gitignore
CMakeLists.txt
ENFT.sln
LICENSE.txt
README.md

README.md

ENFT-SfM

Current version: 1.0

This source code provides a reference implementation for ENFT-SfM.

For ENFT(Efficient Non-consecutive Feature Tracking) method implementation, please go to ENFT.

For Segment-Based Bundle Adjustment implementation, please go to SegmentBA.

1. Introduction

ENFT (Efficient Non-consecutive Feature Tracking) is a feature tracking method which can efficiently match feature point correspondences among one or multiple video sequences. ENFT-SfM is a complete structure-from-motion system which uses ENFT method for feature tracking and SegmentBA for bundle adjustment optimization.

2. Related Publications

[1] Guofeng Zhang*, Haomin Liu, Zilong Dong, Jiaya Jia, Tien-Tsin Wong, and Hujun Bao*. Efficient Non-Consecutive Feature Tracking for Robust Structure-from-Motion. IEEE Transactions on Image Processing, 25(12): 5957 – 5970, 2016. [arXiv report] [video]

[2] Guofeng Zhang, Zilong Dong, Jiaya Jia, Tien-Tsin Wong, and Hujun Bao. Efficient Non-Consecutive Feature Tracking for Structure-from-Motion. European Conference on Computer Vision (ECCV), 2010.

3. License

This software is for non-commercial use only. Any modification based on this work must be open source and prohibited for commercial use.

If you need a closed-source version of ENFT-SfM for commercial purposes, please contact Guofeng Zhang.

If you use this source code for your academic publication, please cite our TIP paper:

@article{
  title={Efficient Non-Consecutive Feature Tracking for Robust Structure-from-Motion},
  author={Guofeng Zhang, Haomin Liu, Zilong Dong, Jiaya Jia, Tien-Tsin Wong, Hujun Bao},
  journal={IEEE Transactions on Image Processing},
  volume = {25},
  number = {12},
  papges = {5957--5970},
  doi = {10.1109/TIP.2016.2607425},
  year={2016}
}

4. Dependencies

5. Usage

The project has been tested in Visual Studio 2015 and 16.04 (GCC Version > 5.4). We provide almost all the prebuilt x64 libraries in3rdparty/. You can run the program directly under the default project setting.

For Windows:

./ENFT.exe path/to/your/config.txt

For Ubuntu 16.04:

sudo apt-get install libx11-dev libglew-dev freeglut3-dev libjpeg-dev libtiff-dev libpng-dev
cd /path/to/the/project
mkdir -p build && cd build
cmake -DCMAKE_BUILD_TYPE=Release ..
make -j4
./runENFT path/to/your/config.txt

Configuration File

We provide three configuration file example in ENFT\config\, which show how to run ENFT-SfM in single sequence mode ( support varying focal length and constant focal length datasets) and multiple sequences mode (the camera intrinsic parameters should be known). Please refer to Dataset part to download these corresponding datasets.

  1. videos_number is the number of videos.

  2. window_width and window_height indicate the size of window.

  3. video_i_*give the information of the i'th video.

  4. calib_file_name is the file name which provide the camera intrinsic parameters (fx,fy,cx,cy). If not provided, the system will assume the focal length is contant but unkown (if const_focal = 1) or varied (if const_focal = 0).

  5. const_focal will be used when calib_file_name not given, set 1 if the focal length of the camera is constant.

  6. radio_distortion set 1 when camera distortion not rectified.

  7. param_directory is the param file directory and it contains the detailed params of tracking and bundle adjustment.

  8. output_directory is the output file directory, it would be the video directory when not given or do not exist.

  9. min_frame_number and max_frame_number is the param of segmentBA for spliting a video to several sequences.

  10. use_temporary_file set 1 to save ACT files and reuse (if exist) temporary binary files (e.g. v0.txt) for some steps in SfM.

  11. view set 1 to show the result.

Dataset

Note

  1. For Windows: To ensure that the program works properly, it is recommended to adjust the graphic card priority in NVidia control pannel, and use the NVidia graphics card (recommend NVidia GTX 780 or above) for ENFT executable program.
  2. For Linux: Currently, the Linux version may be not as efficient as that of the Windows version. It still has problems to run with NVidia driver. So for running the program properly, it is recommended to switch the graphic driver to nouveau. We will try to address this problem in the future.
You can’t perform that action at this time.