Upscaled planetary textures

[john71]

This is a 12k Saturn enhanced map based on Björn Jónsson's Saturn planetary map.

According to his website: "...planetary maps on these pages are publicly available. You do not need a special permission to use them but if you do then please mention their origin in your work, e.g. "created by Björn Jónsson" or something equivalent."

https://bjj.mmedia.is/data/saturn/index.html

Here is the texture:

saturn 12k public.jpg

Added after 3 hours 21 minutes:
Neptune 12k from NASA texture:

https://solarsystem.nasa.gov/resources/2364/neptune-3d-model/

Neptune 12k public.jpg

Added after 2 hours 28 minutes:
Uranus 12k from NASA texture:

Uranus 12k public.jpg

Added after 1 hour 14 minutes:
24k Jupiter from Björn Jónsson's texture (54.7 MB!!!!!! Please only open it, if it is OK for you!):

Image: NASA / JPL-Caltech / SSI / Southwest Research Institute / Malin Space Science Systems / Italian Space Agency (ASI) / Italian National Institute for Astrophysics (INAF) / JIRAM / Björn Jónsson

https://www.planetary.org/articles/0514-juno-meets-cassini-a-new

jupiter public 24k.jpg

[john71]

Venus clouds 12k based on Björn Jónsson's planetary map:

Venus 12k cloud public.jpg

Added after 10 hours 51 minutes:
Pluto 24k from NASA global map:

Credits: NASA/JHUAPL/SwRI

https://www.nasa.gov/image-feature/pluto-global-color-map

pluto 24k public.jpg

Added after 1 hour 12 minutes:
Phobos 24k from Viking NASA texture:

phobos 24k public.jpg

Added after 10 hours 10 minutes:
Charon 24k from NASA/Wikipedia global map:

Charon 24k public.jpg

[john71]

Oberon 8k from NASA texture:

Oberon 8k public.jpg

[fyr02]

Hi,

This is a good attempt, but there are many problems.

Uranus texture was nice. Original texture was fictional, by James Hastings-Trew. Most upscale programs are trained on either generalized algorithm or one specialized on drawings/cartoons (like waifu2x), so it works well with Uranus.

Others no so much. There are many artifacts in the textures. For Phobos, color artifacts ruin the appearance of the texture. Here, it looks more like a bad modern art piece than any good map I would use.

In the case of Oberon, the upscaling has not upscaled the surface features (it still looks blurry), and instead upscaled the RGB noise in the image. It's too easy to overfit with such low-resolution data. Only with a larger volume of data does it become feasible to upscale like this.

I attempted this for Ariel and it turned out not great but still somewhat satisfactory.

It could be possible to also do the denoising using neural networks, such as ReLUNet. It is easy to set up but takes a lot of time to train and to use. My results with it and Oberon were unsatisfactory. The original images are small enough such that separating noise and small surface features is unfeasible by the algorithm.

Also, results are only as good as the training set provided. With many upscalers, the training set is chosen to reflect a broad range of pictures. However, most appropriate training set would be that from a high-resolution mapping spacecraft (MESSENGER, Viking, LRO, etc.), which we can then compare to blurred convolved versions of the same images. That way, the network can learn features associated with planetary bodies.

Of course, making such a dataset is very difficult and time consuming. But doing upscaling without such a dataset is most of the time unsatisfactory.

Finally, for many of these textures, the amount of upscaling is unecessary! For Pluto, 21K map from real data already exists and can be downloaded publicly! For others, it is upscaled too much. i.e, upscaling 3K --> 12K is too much because it will cause the upscaling to be overfit.

So, for all of the textures, it is great attempt, but due to many issues the quality of the upscaled one is only equal or worse than that of the original.

[john71]

Of course you are right - from a certain point of view.

I have to improve them, yes, sure.

But ultimately I have to disagree with you.

My argument is the following:

a.) these are jpg files because I can't use a 250 MB or 500 MB png or tiff file: jpg quality is not the best.

b.) some of the upscaled textures have resolutions like 32k or 64k, but I can't upload them here, because they are huge.

c.) the Jupiter, Saturn and Neptune textures are way better in my opinion, than the originals.

d.) last but not least: these are eye candy textures and they do look better in Celestia than the low resolution versions. Some users can't access any upscaling software, so this can be an introduction for them.

Added after 9 hours 15 minutes:
By the way: if you have a better version, please upload your texture here!

It would be appreciated!

Added after 4 hours 29 minutes:

For Pluto, 21K map from real data already exists and can be downloaded publicly!

Would you be so kind to share the link?

The official NASA texture is only 5k: https://www.nasa.gov/image-feature/pluto-global-color-map

[john71]

James Webb Space Telescope Jupiter upscaled, enhanced 6k:

webbs-jupiter-6k-public.jpg

[SevenSpheres]

Would you be so kind to share the link?

The official NASA texture is only 5k: https://www.nasa.gov/image-feature/pluto-global-color-map

It's a grayscale map, which can be colorized based on the lower-resolution color map or photos: https://astrogeology.usgs.gov/search/map/Pluto/Ne ... ons_Global_Mosaic_300m_Jul2017

[john71]

Thanx!

I can make an 80k texture from it!

This is a 20k BW upscaled part of the whole map:

Sorry I can't upload it. The Forum web UI won't add it.

[Gurren Lagann]

Well you could start using zip/rar/7z compressed files.

[john71]

You are a genius.

It worked.

But there is some kind of problem with the Forum in my opinion.

Someone should check it.

Pluto80kpublic.7z

[john71]

JWST HIP 65426 b upscaled and enhanced:

HIP 65426 b public.jpg

[john71]

Sun's entire surface as seen in extreme ultraviolet light (304 angstroms) for the time period Jan 1 - Sep 27, 2012.

Credit: NASA JPL

Source:

https://svs.gsfc.nasa.gov/30362

Upscaled to 16k:

euvi_aia304_2012_carrington 16k small PUBLIC.jpg

[john71]

Eros 8k texture from NASA 3D model texture:

https://solarsystem.nasa.gov/resources/2389/eros-3d-model/

Eros 8k small PUBLIC.jpg

[Sirius_Alpha]

Why do you think it looks so different from, for example, this texture-map of Eros?

[john71]

I don't know, ask NASA. Not my texture. Hint: maybe it is not a sphere.

[Sirius_Alpha]

ask NASA. Not my texture

Wait where did you get the "Eros 8K texture" that you posted?

[john71]

It is the texture of the official NASA 3D model.

[Sirius_Alpha]

I have no idea where the texture is from, to be honest. But the point wasn't to try to "upscale" another texture, but rather to inquire about your process. One would think that an 8K texture of Eros would look like Eros. The fact that your texture deviates so significantly from what Eros is known to look like would, for anyone, cause wonder of how you got that texture, and whether the methodology is reliable.

[john71]

My methodology is the following: I upscale and/or color NASA or other legally usable free planetary textures.

So I had to delete the upscaled version of your texture.

[Sirius_Alpha]

So why doesn't your upscaled texture look like the asteroid it claims to represent? You upscaled the texture I posted and it looked like the same asteroid, for example.