9 PEP2026 Tutorial

Access Virtual Desktop (if needed)

If you are using browser-based computing resources, you can use your guest credentials and instructions at HCC OnDemand - HCC-DOCS to log in and access your virtual desktop. Note that guest credentials will not use Duo authentication. Once logged in, proceed to the HCC OnDemand directions for starting up a virtual desktop.

When requesting your virtual desktop resources, enter the following settings:

• Request RAM in GB: 16 GB
• Partition selection = gpu
• GRES = gpu

Once the virtual desktop is launched, open the terminal (terminal emulator) and enter the following commands:

• module purge
• module load grime-ai
• grime-ai

After a few seconds, you should see text appearing in the terminal window. After perhaps a minute, you should see the GRIME AI interface appear.

Try these steps to test out GRIME AI

1. Try downloading images from NEON/Phenocam Sites.
2. Load your downloaded images in GRIME AI.
3. Create a Composite Slice image.
4. **[Optional]** Tune a SAM2 Model.

How to tune a SAM2 model with GRIME AI

First, if using a virtual desktop, look in ~/Documents/GRIME-AI/SAMPLE_DATASETS to find a sample model training dataset. Or, if on your own computer, download a sample model training dataset. Unzip the sample training data files, noting the location (file path) of the uncompressed folder.

Click the Deep Learning ("brain") icon on the GRIME AI toolbar.

Use the browse button to navigate to the folder containing the uncompressed training data folder.

Move the training data folder into the Selected Images Folders window. Be sure to select "Staff" from the dropdown label selector. For the first run, we suggest choosing <12 epochs in case model training is slow on your computer. Then, to train a model, click the "Train" button.

After a few moments, a status wheel should appear. Watch the debug terminal (black screen) to see the modeling process proceed. Please send us any errors you encounter!

Once the model is complete, you should find it in the Documents folder on your computer. There will also be model diagnostic plots and subfolders with example model outputs (image overlays). Check to see if the model was successful in identifying the staff gauge in these images! An example directory path is "C:\Users\computerusername\Documents\GRIME-AI\Models\sam2\20260121_114911_TEST_01".

More details about the model training and deployment can be found in GRIME-AI Model Building and GRIME-AI Model Deployment .