# Framework for "Fast-Point-Feature Histograms" ## Dependencies The code is tested under Ubuntu 12.04, Ubuntu 14.04, and Ubuntu 18.04. It depends on the following packages: * cmake (>= 2.6) * libeigen (>= 3.0) * libboost (>= 1.46) * libqt4-dev (>=4.8.7) The following command should install all needed dependencies: Ubuntu 12.04/14.04/18.04: $ sudo apt-get install build-essential cmake libeigen3-dev libboost-all-dev libqt4-dev We tested this on Ubuntu 12.04.05 LTS, Ubuntu 14.04.02 LTS, and Ubuntu 18.04.3 LTS, but it should also work for versions in between. ## Setup of your project and compilation In the following, we assume that your are already in the directory with your working copy of the directory. 1. Execute the setup script "setup.sh", which creates the "build" directory: $ sh setup.sh The script performs the following steps: (a) Create a "build" directory, which is used to store the object files, etc. (b) Calls "cmake .." from the "build" directory to generate the Makefile (c) Calls "make" inside the build directory to compile the framework. (d) Calls "./runtests" in the base directory of the framework. There should be a "data" folder containing the training and test data for this project and a "build" directory in your working copy. You should not commit the build folder to the repository, since it can be recreated with the setup provided script. You only have to commit your changes to the source files and your final documentation. All executables and temporary build files are only needed locally on your machine. 2. Execute "./runtests", which should run the provided tests: $ ./runtests Due to the build up of the test cases, those for FPFH, SPFH and Features do not fail as long as the set of neighbors is always empty. Only if your k-d tree radius neighbor search works, these test cases will start failing! For grading purposes, we will use the same tests to check if you provided a working solution of the subtasks, but you will not get points for these tests by passing them using empty neighbor sets. ## Notes on the provided framework Your code and implementation should go into the provided classes in the sub-folder "project". Your final solution should complete the methods with missing code. For some tasks, we provide "test cases", which should pass if your code solves the subtask. Parameters can be entered using the command line (in eclipse right click on project -> Run Configuations -> Arguments). Type -help to see a list of parameters and a description. The standard parameters are provided at the beginning of the main method. The training and test data is expected to be in the data folder. This folder includes a subdirectory "test", containing all files currently used for testing, and a subdirectory "train", currently used for training the SVM. The subdirectory "histograms" contains the histogram feature files created for the test and training data. For the post-processing import to work, you have to specify the subdirectories "model" and "predicition" to be the output destinations for the svm-light model and predicition files. The subdirectory "post_processed" contains the final label files, after performing region merging. After compiling the code, you should find four executables in the root directory: 1. "create-histograms" is used to create histograms for training and test data and label them as "berries" or "stem". 2. "classify-histograms" is then used to learn a model using svm-light and classify the test histograms based on this model into berries and stems. 3. "post-processing" applies a region merging algorithm to the classified test set to provide a unique labeling to connected regions. 4. "evaluate-files" gives you the possibility to compute the final precision and recall values from the post-processed files.