So begins the Python Computational Geomorphology Software System!

Steps

1. Use the GEE Code Editor and gee_scripts to grab ya imagery
2. Use make_jpges.py to turn ya Earth Engine files into jpgs for Doodler.
3. Use Dash Doodler to label/segment ya imagery
4. Use Segmentation Gym to segment unlabeled imagery

Plan

Perhaps the only time in history someone is putting a figure from an in-prep NSF proposal so as to say "Please, if you can do this instead of me, go ahead"

Guides

General Python guides

I have been compiling a guide to Python-based landscape analysis on my Notion site, but with the advent of PyCoGSS, I anticipate moving the guide to this repo.

Doodleverse guides for Dartmouth/HPC

This is a guide for accessing Doodleverse software in Dartmouth's HPC environment. Please consult the wikis of the Doodleverse software for help there.

Doodler

You have to make an ssh tunnel to copy and paste the address that Doodler spits out:

1. ssh to andes or polaris
2. Do the instructions on the Github to install dash doolder. When installing Dash Doodler navigate to the dash_doodler directory with cd dash_doodler. Don’t worry about conda version errors. It will take a while.
3. activate your dashdoodler environment you created when installing Doodler
4. Run python doodler.py after navigating to the directory with cd dash_doodler
5. You have to make an ssh tunnel to copy and paste the address that Doodler spits out. Note the address: it should be http://127.0.0.1:8050/ The server is 127.0.0.1 and the port is 8050.

6. Open another Terminal. Do ssh like you normally would but add flags -NfL (note capitals). The general syntax for ssh tunneling is:

ssh -NfL [port]:[server]:[port] [your net id]@[linux machine address]

so replicate the following but with your username

7. Copy and paste or type the http address into your web browser of choice. It should pop us as Doodler!!

Segmentation Gym

1. Use the utils/gen_images_and_labels.py script in Doodler (in your Doodler environment!) to generate the labels and images folders that you can point segmentation gym to. It helps if you put all your results that you want to use in one directory. A dumb thing is that you have to click into the directory when the prompt comes up (rather than selecting the folder).
2. Log into Discovery

ssh [id]@discovery7.dartmouth.edu

3. Start an interactive session on Discovery. when you type k80:4 that means you are using 4 GPUs at once 😎 you do training in super-speed!! But if you aren’t doing any operations it could time out because those resources are sometimes high in demand.

srun -p gpuq --gres=gpu:k80:4 --pty /bin/bash

4. Making sure you installed Segmentation Gym according to the GitHub, activate your gym environment

conda activate gym

5. Follow the Segmentation Gym wiki and have fun!