Mesas 042 experiments

Datasets and code to replicate experiments of Halodova et al: Adaptive image processing methods for outdoor autonomous vehicles

In this section, we present how to execute scripts to replicate the experiments performed.

Datasets and code to replicate adaptation experiments:

1. Download datasets to some directory.
2. Get the code, i.e. experiments_2018_mesas_exposure branch branch from github.
3. Compile the code, source the workspace, run roscore.
4. Edit the script process.sh in the directory /stroll_bearnav/mesas_2018_exposure/. If you want to replicate all experiments, make sure that all if statements are true, resp. 1 == 1. These statements switch on or off the individual experiments described in the paper.
5. Go to the directory /stroll_bearnav.
6. Execute the script process.sh, which is stored in a folder /mesas_2018_exposure. The first argument is the dataset directory. Make sure it is without the last slash. For example ./mesas_2018_exposure/process.sh ~/experiments-mesas-exposure
7. Results (also graphs) can be found in the created folder results.

Datasets and code to replicate the prediction experiment:

1. Make sure that you run the relevant experiments from the previous section, especially that the "Summary map" is created (see the ./mesas_2018_exposure/process.sh.
2. Get the code, i.e. experiments_2018_mesas_exposure_predictions branch branch from github.
3. Compile the code and source the workspace.
4. Edit the launch/evaluate-predictions.launch to set a proper folder (the temporal folder with the 87 maps).
5. Go to the directory /stroll_bearnav.
6. Execute ./mesas_2018_exposure/build_temporal_statistics.sh stat.txt to build feature visibility statistics.
7. Execute ./mesas_2018_exposure/test_temporal_models.sh stat.txt to build feature visibility statistics.
8. Copy the resulting FreMen_X files to the results folder.
9. Rerun the ./mesas_2018_exposure/evaluate.sh script for the statistical tests and graph generation.

Experiment setup overview

References

1. L.Halodova, E.Dvorakova, F.Majer, J.Ulrich, T.Vintr, K.Kusumam, T.Krajnik: Adaptive image processing methods for outdoor autonomous vehicles. In Modelling and Simulation for Autonomous Systems (MESAS), 2018, in review. [bibtex]