MODD evaluation and processing package, Release 25082016 Authors: Matej Kristan, matej.kristan@fri.uni-lj.si (processing code, ./accv) Janez Pers, janez.pers@fe.uni-lj.si (evaluation and dataset code) Rok Mandeljc, rok.mandeljc@fe.uni-lj.si (cleanup and coder adaptation) Vildana Sulic Kenk, vildana.sulic@fe.uni-lj.si (cleanup and coder adaptation) If you use the code or reference the results, please cite: [1] M. Kristan, V. Sulic Kenk, S. Kovacic and J. Perš, "Fast Image-Based Obstacle Detection From Unmanned Surface Vehicles," in IEEE Transactions on Cybernetics, vol. 46, no. 3, pp. 641-654, March 2016. doi: 10.1109/TCYB.2015.2412251 [2] Matej Kristan, Janez Pers, Vildana Sulic, Stanislav Kovacic: A Graphical Model for Rapid Obstacle Image-Map Estimation from Unmanned Surface Vehicles. ACCV (2) 2014: 391-406 The code is complete evaluation and processing (algorithm) package. Only separate download you need is the MODD dataset, which can you download here. http://vision.fe.uni-lj.si/RESEARCH/modd/ Please note that the code has been tested to be working with Matlab R2015b x64 under Linux (CentOS 7). For the verification, we provide folder reference_results which contains .mat file and Excel file, produced by the code, so you can test if your setup works ok. We do not make any other guarantees about the code quality - it is provided as is. Requirements for running the code: 1) MODD dataset (see above). 2) ffmpeg installed somewhere in your path, to produce videos for visual evaluation - otherwise you will get only the numerical results. Installation and running: 1) Unpack the code. 2) Download and unpack the MODD dataset into the ./dataset folder that has been provided with this code (but it is initially empty). Directly under ./dataset should be dataset video folders, e.g. ./dataset/01, etc. 3) (optionally) install the ffmpeg for videos 4) move into the root of the unpacked code and run water_dataset_fullevaluation.m 5) In the ./results folder there will be output videos and the .mat file with long and complicated file name, reflecting the parameters used. 6) on the .mat file name, run the water_dataset_analyze.m with the argument being the previously created .mat file. This only works under Windows, as it produces Excel file. Excel file will be created. 7) (on first run, with unchanged code) compare the Excel file created with the file in ./reference_results to see that the code is running ok. Changing parameters: 8) You can edit the file water_dataset_fullevaluation.m and produce results for different videos from the MODD dataset, with different parameters. Evaluate your own method: 9) Please see the function water_dataset_evaluate.m and insert call to your own code instead of the lines: detector = water_perform_detection_on_image(detector, im); We are unable to provide detailed documentation for the detector interface, but since we provide the full source it should be easy to examine the interface and duplicate it for your own algorithm to compare it with ours.