UMD (Unified Maritime Dataset)

This dataset accompanies the paper A Multimodal Dataset for Autonomous Surface Vehicle Situational Awareness. It aims to collect the marine datasets available for surface obstacle detection and tracking tasks, and provide annotations for them. The annotations are presented in JSON format. An API for accessing and editing the data, as well as reproducing the tests proposed in the paper, is given at https://bitbucket.org/isme_robotics/pods/src/main/ (branch name: paper). Before using the scripts, remember to modify the file cfg.yaml, changing the dataset path with the path of the folder you downloaded the dataset in. In the following, the mapping from the dataset IDs reported in the paper and the actual dataset folders is reported. RGB1 (SeaShips 7000 RGB) -> rgb/SeaShips_scenesRGB2 (Singapore Maritime Dataset Plus) -> rgb/smd_plusRGB3 (SMDP) -> spscd_scenes (sequences with obstacles), spscd_bg_scenes (sequences with no obstacles)RGB4 (Pep Buoys) -> rgb/RF_boats RGB5 (Lentic waters dataset) -> rgb/RF_boats RGB6 (Crane finder dataset) -> rgb/RF_cranesRGB7 (Sailboats dataset) -> rgb/RF_boats (the non-sequential frames), rgb/RF_sailboats_scenes (the sequential frames)nesRGB New data -> rgb/new (sequences with obstacles), rgb/new_bg (sequences with obstacles) IR1 (Singapore Maritime Dataset) -> ir/smd_ir_scenesIR2 (Raytron) -> ir/raytron (frames with real boats), raytron_mockup (frames with only mockup boats) MLT1 (MIT Grant) -> mlt (original dataset at https://seagrant.mit.edu/auvlab-datasets-marine-perception-2-3/)