Coordinating repo for ESGF + Climate in Society + AI Foundry (ANL, UC, NIU)
One can find important notes/references which we used throughout the project in /references
Porgress Updates can be found in /updates
Jupyter notebooks containing experimentation and results can be found in /notebooks
The notebooks are listed below:
low-resolution-images-to-high-resolution.ipynb
- Contains our initial experimentation with a basic UNet Model.
- We also tested this out with an SRCNN model and found the UNet Model to work better.
- Running the notebook with the proper data outputs relatively good, but not great, results.
16-to-32-model.ipynb
- This notebook contains the final training and evaluation of a model which goes from 16x16 resolution to 32x32 resolution (pixels)
- This was one of the 3 models in our stacking experimentation, where we stepped up resolution scales, similar to the DeepSD paper.
8-to-16-model.ipynb
- This notebook contains the final training and evaluation of a model which goes from 8x8 resolution to 16x16 resolution (pixels)
- This was one of the 3 models in our stacking experimentation, where we stepped up resolution scales, similar to the DeepSD paper.
5-to-8-model.ipynb
- This notebook contains the final training and evaluation of a model which goes from 5x5 resolution to 8x8 resolution (pixels)
- This was one of the 3 models in our stacking experimentation, where we stepped up resolution scales, similar to the DeepSD paper.
combining-models.ipynb
- This notebook combines all the models from the previous 3 notebooks into one model, and runs evaluation on the dataset
- We find this to give very good results (much better than going straight from 5 to 32)
- https://arxiv.org/pdf/1703.03126.pdf
- Used for the basic structure of the stacking model
- https://arxiv.org/pdf/1501.00092.pdf
- Used for benchmarking our results
- https://openaccess.thecvf.com/content_ICCV_2017/papers/Tong_Image_Super-Resolution_Using_ICCV_2017_paper.pdf
- Used for benchmarking our results