CUDA Implementation of Killing Fusion, from the paper
For a short intro about Killing Fusion and its demo, see slides
To build the code, from the repository folder perform the following steps:
mkdir build cd build cmake all -DEIGEN_INCLUDE_DIR="./../third_party/include/eigen3" .. make
For running the code:
Download, for example, the Snoopy dataset: http://campar.in.tum.de/personal/slavcheva/deformable-dataset/data/Snoopy.zip. Extract the dataset to /data/snoopy folder.
Download the camera intrinsics parameters for the Snoopy dataset from : http://campar.in.tum.de/personal/slavcheva/deformable-dataset/data/intrinsics_kinect1.txt Place the camera intrinsics file in /data/snoopy
Create a result folder at /build/bin/result to save the final meshes. It has to be this path, because it is hardcoded for now.
Finally run the optimization:
./bin/killingfusion -i ../data/hand/ -b 2 -f 22 -n 1000 -wk 0.1 -ws 0.05 -a 0.1
Details about the input flags -i input dataset path -b first frame index, which is the canonical model -f last frame index of the sequence -n max number of Gradient Descent steps -wk killing term weight -ws level set term weight -a parameter to control rigidness in deformation field -d activates debug mode, which saves intermediate data and prints energies