jiwoncpark / aracle Goto Github PK

PIL doesn't seem to be pasting images as expected.

The directory structure we have in mind is
helio-ai
---data/
------generate_Y_segmented_maps.py
------generate_Y_annotations.py
------data_utils/
---------download_sharp_data.py
---------visualize_data.py
where generate_Y_segmented_maps.py will import modules from download_sharp_data.py.

• Separate out scripts/modules
• Add numpydoc docstring into all the scripts/modules
• Refactor to get a function in download_sharp_data.py that fetches the SHARP data and another that creates the Y_metadata Pandas dataframe.
• Add script for generating the dictionary, generate_Y_annotation.py.
• Delete extraneous notebooks
• Separate out video generating function and place it in visualize_data.py.

With RTD

• Mathieu et al 2016 showed that loss based on GANs produce sharper generations than do L2-based losses. (torch implementation here)
• Denton and Fergus 2018 unsupervised video generation model that learns a prior model of uncertainty in a given environment. Video frames are generated by drawing samples from this prior and combining them with a deterministic estimate of the future frame. Relevant because we do want to assign different priors to regions near and far from the equator.
• Reda et al 2018 present a both past frames and optical flows.