Scene reconstruction with fast fusion

In order to visually evaluate the precision of PUTSLAM estimate it is possible to use FastFusion software created by F. Steinbruecker, J. Sturm and D. Cremers.

First, we need to download the software from git. Go to the wanted directory and then write:

git clone https://github.com/tum-vision/fastfusion.git

When finished write:

cd fastfusion && cmake . && make

If everything goes ok, the fastfusion system is ready to be used. We will need the file with trajectory, e.g. graph_trajectory.res, that is created everytime the PUTSLAM system is executed. We can also use the trajectory from best results.

The exemplary execution will be perfromed on matching result for Freiburg1_desk link. Extract the zip file with best results and copy the graph_trajectory in PUTSLAM/scripts:

unzip freiburg1_desk_matching.zip 
cp freiburg1_desk_matching/bestResult/graph_trajectory.res YOUR_PATH/PUTSLAM/scripts

We need to convert the file with the results to the format compatible with fastfusion. It can be done with prepared script and store it in fastfusion.txt:

cd YOUR_PATH/PUTSLAM/scripts
python createFastFusionFile.py graph_trajectory.res > fastfusion.txt 

Now, we need to copy fastfusion.txt file to directory with freiburg1_desk:

cp fastfusion.txt YOUR_PATH/Datasets/rgbd_dataset_freiburg1_desk/
cd YOUR_PATH/Datasets/rgbd_dataset_freiburg1_desk/ 

The fastfusion can be run with command:

RELATIVE_PATH/fastfusion/bin/onlinefusion fastfusion.txt --thread-fusion 

To start press 'S'. You can zoom in/out with CTRL + mosue wheel, you move screen with right mouse button and you rotate the view with left mouse button. Have fun!

Important! When using datasets recorded from messor, quadcopter or Kin1vsKin2, you should additionaly inform fastfusion about depth image scale with: --imagescale 5000

Example: RELATIVE_PATH/fastfusion/bin/onlinefusion fastfusion.txt --imagescale 5000 --thread-fusion