Using the GPU server - HPI-DeepLearning-PlaceRecognizer-SS17/fetch-flickr GitHub Wiki

First, ssh to the machine (only possible via VPN or from HPI). On UNIX systems you can also pass the -X flag to be able to run programs with an GUI, like gedit.

ssh [email protected]

You can change to a virtual environment using workon:

workon env_anotherTry

You can use this environment if you don't make any big changes, mxnet-ssd is already running there. If you want to create your own you can run

mkvirtualenv -p [2,3] my_new_env

In your new environment you can install all kind of python stuff and it will be separated from the rest. -p2 if you want to use python 2 as default (necesseray for the mxnet-ssd repo code, dunno what normal mxnet wants, I think python 3?) There is a requirements.txt in mxnet-ssd which installs the stuff needed for that repo.

There are scripts to prepare the building training tools/prepare_allbuildings.sh and to start the building training train_buildings.sh If you want to train on a new dataset, these can easily be adjusted, aswell as the prepare_dataset.py and labelme.py (only needed to change the classnames I think)