This is my research code for the paper "A Learning Approach for Real-Time Temporal Scene Flow Estimation from LIDAR Data" presented at ICRA 2017. Please contact me (aushani@gmail.com) with any questions or comments. Please cite the paper if this code is useful in your work. (bibtex)
If you have any questions, please contact aushani@gmail.com. Please note that the models here (background filter, occupancy constancy) are trained for the KITTI dataset.
Cuda is assumed to be already installed (you need a GPU to run the code). Other dependencies are bazel, Qt4, and OpenSceneGraph. You also need the KITTI data. The bootstrap.sh script takes care of these except for cuda.
To get started, run:
$ ./bootstrap.shYou will be prompted for:
-
Bazel 0.4.5
Build system to compile the project. The project was originally built with bazel 0.4.5, and I have received reports of failures when compiling when more recent versions of bazel.
-
Qt4
from apt-get
-
OpeneSceneGraph
Used for rendering, locally built
-
KITTI Data
Prepackaged in the directory structure that the code is expecting. You also have the option of downloading a small snippet of it if bandwidth or diskspace is a concern.
To compile everything, go to the cc directory and build with bazel:
$ cd cc
$ bazel build //...After building, you can run with a command like:
$ bazel-bin/app/flow/viewerThis assumes that the data was put in ~/data/tsf_data. Otherwise:
$ bazel-bin/app/flow/viewer --tsf-data-dir path/to/tsf_data