Full training data

[fommil]

Hi, I'm really interested in recreating this model in pytorch (or similar) to produce an onnx model that I can run in the cloud, because my machine doesn't have the compute. Fully credited of course. But your training data set in github only has one file on it, is this what you used or do you have a larger data set available? I can't imagine you only used one file.

I remember back in the 1990s sitting at IRAF for hours on Hubble data manually picking the best PSF for each star to create colour/magnitude diagrams. I don't know where all those fits went, they might be available as part of public science data sets. But I've been wondering how useful that would be as a training data set for a more modern CNN model like this.

Thanks for building this model and making it available, it's like magic!

[fommil]

Ah, sorry, I just read the section "Training Dataset" where you say you want to keep it to yourself. I didn't see this is the training files, it's only in the README. That's a real shame.

[nekitmm]

You can find different versions of the weights on the starnet website. This version, for example, has torch weights in it: https://www.starnetastro.com/experimental/

[fommil]

thanks @nekitmm. I'm not particularly interested in weights though, I'm interested in the theory and datasets behind the code, so I was a little disappointed to find out that the training data (or the process for creating training data) isn't publicly available, but I respect your decision to keep it private.