Find useful data and/or case study

[griffin-pitts]

In order to begin concrete implementation of top image segmentation algorithms, a dataset is needed.

While smaller-sized examples of algorithms can currently be created and used, future versions will need to be fine-tuned for petrographic-like images, i.e, the same model that has been trained on a cats/dogs dataset will not accurately help predict data within smaller-sized petrographic data.

This initial data set should contain quality images with compact structures, like the image below.

Better than a basic petrographic dataset, a case study containing like-structured images, would be useful to tell a story within a publication.

This all depends on what data is available.

[griffin-pitts]

https://arxiv.org/abs/2204.05114 - "Generative art" to create petrography dataset

[nick-gauthier]

https://arxiv.org/abs/2204.05114 - "Generative art" to create petrography dataset

Great idea, I was just going to suggest something like this.

[nick-gauthier]

Running list of potential data sources on the wiki. Many of these are exactly what we're looking for (paired thin section images in both PPL and XPL with some kind of annotation) but not all are easily downloadable. Depending on the data permissions for each, we may be able to automatically scrape the images or just screenshot what we need.