Quote:
Originally Posted by strongmanmike
Great news and interesting write up Mat, why make a single discovery  ... when you can make bloody nine!   Well done to the team  ...and as an extra bonus, you have produced a unique, quite beautiful, not to mention deep, image of an an otherwise rarely shot bit of sky. Now you need to team up with the other ultra deep Kiwi guru, Rolf Olsen, one can only imagine what the pair of you might discover together  
Slightly off topic but still relevant?...Your processing note is interesting too. In this new age of AI sharpening software/plugins, it seems to matter which software/plugin you actually use and the technique applied. While in this case your discovery is not about the fine details revealed anyway but rather the existence of the whole object and I am not an expert but in relation to many other images, enhancing fine details, based on learning what astronomical detail looks like in lots of other images, seems fraught with issues...  eg. what if the neural networks learn from lots of poorly enhanced details in some/many of those images?
Mike |
Thanks Mike! the project kept us very busy 😀 As you mention, this patch of sky has been quite overlooked by amateurs and also professionals, which certainly helped in making the new discoveries. I think the very dense star field is a big factor in this, and it created its challenges with processing. Working with Rolf would be great! He is a major inspiration for my imaging, and his deep images are just fantastic.
Discussing the role of AI in astro processing can seem a bit like opening pandoras box 😂 For sharpening and noise reduction afaik there is the Xterminator range (which I understand was trained on Hubble data that was convolved, so it represents a best case scenario as far as detail) and Topaz. One criticism of Topaz is it is mainly trained with terrestrial images but in both cases it’s a bit of a black box to most of us as to exactly what the neural network is doing. Like all tools I think you can go too far with them and create artefacts, but if you use them carefully and compare back and forwards with your raw data to make sure its enhancing what you have captured, I think they are useful.