The DL4DS module has been fully deployed at Gadi under the project dk92. Please join project dk92 to run it. 

The original DL4DS tutorial notebook has been revised to make it runnable at Gadi.  Please clone it from  https://github.com/nci/DL4DS-nci  to Gadi. 

You can start an ARE JupyterLab session to run it. 

MAELSTROM Dataset

This notebook shows how to train the DL4DS using the MAELSTROM dataset. Detail about the dataset can be found in the following link: https://git.ecmwf.int/projects/MLFET/repos/maelstrom-downscaling-ap5/browse


Launch an ARE Instance

We have deployed DL4DS on the ARE, to launch an instance do the following:

  1. Go to https://are.nci.org.au/ and log in. 
  2. In the JupyterLab launch Dashboard, use the following information:
    Walltime (hours):4
    Queue: gpuvolta
    Compute Size: 1gpu
    Project: your own project (such as ab123).
    Storage: gdata/dk92+scratch/ab123
  3. In the Advanced options section: 
    Module directories: /g/data/dk92/apps/Modules/modulefiles 
    Modules: dk4ds/1.8.0
  4. Hit the launch button

Launch the Notebook

  1. In the JupyterLab, click view → open from path and enter the location containing the cloned notebook as above.  
  2. All, codes are already present, no need to write anything.
    (If you want to run the cells, must start from the top)

Inference Results for the MAELSTROM Dataset.

The following video shows an example of the DL4DS model prediction, in this case, the MAELSTROM dataset is used in training. Left Hand side is the target image, right hand side is the output of DL4DS method. On the top left hand side, a slider is used to select different timeframes. The result shows that prediction is really close to target downscaled image. 


































  • No labels