On this page, you will be able to download the parts of the dataset which interest you. We also provide overview samples for each training environment and trajectory.
Warning: We patched hdf5 files of the dataset on the 6th Sep. 2019. If you downloaded some of them before this date, you should download them again to apply the update.
Due to the sheer size of our dataset, we had to fragment it in several individual compressed archives and let you chose which one you want to download. Our download procedure works as follow:
Once the form is submitted, the website will generate a text file containing all links to the archives containing the requested data. Archives can then be downloaded using
wget as follows:
wget --content-disposition -x -nH -i path_to/download_config.txt
This command line will take care of downloading all the archives whose link is in the file and of storing them in the correct directory (in order to respect paths encoded in the hdf5 files). Once all archives are downloaded they can just be uncompressed in place. This can be done by running the following command line at the root of the dataset:
find . -name "*.zip" | while read filename; do unzip -o -d $(dirname "$filename") "$filename"; done;
In case of need, we provide detailed information about our dataset organization here.
Choose the type of visual sensors for which you want to download the records. If you don't select any, you will download only the hdf5 files corresponding the the climate setups chosen at the next step.
Please note that you explicitly need to select every stream you want (e.g. for semantic segmentation tasks, you need to select both the semantic segmentation maps and the left RGB captures). Details on the correspondence between visual maps and RGB captures are given here.
Choose the trajectories and the climate setups for which you want to download visual sensor records. Complete corresponding hdf5 file will be downloaded when at least one trajectory is selected for a given weather setup.
Volume of data selected : 0