In this use case we study the model performance of a building segmentation task using different satellite imagery data products: one processed on the ground and the other processed on board the satellite.
This use case was presented at the 8th International Workshop on On-board Payload Data Compression https://atpi.eventsair.com/obpdc2022/.
Use docker and nvidia-docker as follows:
- Build the docker image
Check the variables in the Makefile, then
make build- Raise a running container
make container-
As soon as the docker is running in background you can launch the following services whithin it:
-
Launch a jupyter lab server
make nb
Then access it from your browser by using this address
localhost:8088/?token=commodities -
Stops the jupyter server
make nbstop
-
Launch an mlflow server
make mlf
Then access it from your browser by using this address
localhost:5055 -
To raise an interactive shell from our running container
make execsh
-
To run tests
make test -
Stop the docker container and everything that is running within it
make stop
-
-
Run the experiment
Inside the docker run
python iqf-usecase.pyuse the -h option to see the possible options for this script.