Skip to main content

Open Drone Map

Open Drone Map (ODM) can be run on a single Palmetto 2 node with Apptainer. For certain stages of the pipeline, a GPU can speed up execution. Since the operation only uses the GPU for a short time, please request a GPU that is less in demand, such as the P100 GPU.

Run ODM interactively​

For small processing jobs, you can run ODM in an interactive job. The salloc command below will request 14 CPU cores, 60GB of memory and 1 GPU for 2 hours.

salloc -n1 -c14 --mem=60G --gpus=p100:1 -t 2:00:00

For this example, we will be downloading an example dataset from the ODM GitHub. The images in this project contain GPS coordinates to be used by ODM.

Setup the ODM directory structure like the following:

mkdir /scratch/$USER/odm
cd /scratch/$USER/odm
git clone https://github.com/OpenDroneMap/odm_data_waterbury.git

When running ODM, it will assume the images are located in datasets_dir/project_name/images/ where datasets_dirin this case is /scratch/$USER/odm.

Finally we can start the processing. Use Apptainer to launch the program.

apptainer run -B /scratch/$USER/odm/:/datasets --nv docker://opendronemap/odm:gpu --project-path /datasets odm_data_waterbury --max-concurrency $SLURM_CPUS_ON_NODE

In the above command, we mount /scratch/$USER/odm to /datasets inside the container. This step is completely optional as you can specify the --project-path argument to ODM to whatever your directory is. We also use the opendronemap/odm:gpu container image since we requested a GPU. If not using a GPU, use the opendronemap/odm:latest container image. Finally, the --max-concurrency flag passed to ODM is set to $SLURM_CPUS_ON_NODE. This is needed to limit ODM to only using the number of threads as CPUs requested for the job. In the above example, $SLURM_CPUS_ON_NODE would be 14.