Results file structure

The files containing the results of the execution of an algorithm on the benchmark are structured as follows:

• A single file for each sequence, named <sequence_name>_[local|global]_result.txt. For example: gazebo_summer_local_result.txt. The sequences' names are the same of the problems files in the Point Clouds Registration Benchmark.
• Each *_result.txt file is composed of:

#<command used to launch the experiment>
#<parameters>
id; initial_error; final_error; transformation; status_code
<id>; <init_error>; <final_error>; <t0> <t1> <t2> <t3> <t4> <t5> <t6> <t7> <t8> <t9> <t10> <t11> <t12> <t13> <t14> <t15>; <ok | oom | runtime_error>

The first two lines allow to reproduce the same tests we originally used to produce the results. They indeed contain all the parameters and, if necessary, the pre-trained model used. <id> is the id of the registration problem in the Benchmark. <initial_error> is the distance between the source point cloud, after applying the initial misalignment and the source point cloud after applying the ground truth. The distance is calculated as defined in the Point Clouds Registration Benchmark. <final_error> is the same of <initial_error> but is calculated using the transformation estimated by the algorithm and the ground truth. <t0>...<t15> are the elements of the estimated transformation matrix. This 4x4 matrix is composed as following: t0 t1 t2 t3 t4 t5 t6 t7 t8 t9 t10 t11 t12 t13 t14 t15

The status_code is useful to note if the algorithm succeeded, if it crashed because it required too much GPU memory or if it threw a runtime error for other reasons.

[Results file example](https://github.com/iralabdisco/neural_registration_comparison/blob/1bdbdf4c31726163233f1aa37eede4bc13b4aabc/results/example_result.txt)