Refining and classifying 2,500+ Hubble images with unsupervised clustering methodologies for the publication of a refined AI model training asset to be used in future GAN development.
Many of these images are absolutely spectacular, but many of also suffer from deformations, pixelations, blur etc. What’s crazy is even some of the really messed up ones take your breath away. I have never craved something visual until this. Images of space arn’t just beautiful , they are inspiring, and all the money that goes into taking those photos with that big expensive satellite, we really should make the most out of every image we can. Not to mention the JWST is making its way to its new home out in space, just imagine what a honed image enhancing model will reveal then!
this dataset is 88 gb processing the raw files into something more usable, and clustering them are going to be computationally expensive and we hope to add value to the community right way by getting this data improved and published on our road to Creating a Hubble Gan models in the future.
- Publish full dataset
- Annotate each image with its cluster id
- Open source documentation on unsupervised clustering methods
- Create a sample grid of images, one image from each cluster to be published as the ocean dataset sample image
- attempt to restore images which look incomplete by using zero-shot in-painting approaches
Lead: Timothy L Carter
Background: Data Scientist & Program Coordinator
DIrectory : Notion – The all-in-one workspace for your notes, tasks, wikis, and databases.
Aide: Pushkar Paranjpe
Background: Data Scientist & Machine Learning Engineer
Note: I am in the directory but could not determine how to link to my entry. Pushkar is a working member on the Athena Project.