Hi Shawn, you might want to try out our new denoising AI model 2.0.0 that was released just a few hours ago. It has to be used in combination with the new GraXpert v3.0.2. We found that it performs better on dark nebulae than the previous version and it produces less artifacts. I have to disagree that SNR is a good measure for how well the denoising is performing. SNR can measure how much noise is left, but it does not measure how close the denoised image is to the real object without noise. I suggest another experiment: Produce.a stack of an object with 1 hour of exposure time, and a stack of the same object taken with the same equipment with an exposure time of 10 hours. Then remove gradients in both stacks, register them and fit them on each other. The 10 hour stack should then be a much less noisy version of the 1 hour stack. Then denoise the 1 hour stack and measure the difference of the result to the 10 hour stack.
@steffen1181 3.0.2 works great. Thanks a lot for the HW Accel Switch. I'm one of those, who had issues with the Denois. Did you use also images from the Seestar for training the model ? I think, there is a big big difference to images from other equipment and that should be considered appropriately.
@@Thomas-mn5ih I don't think so. There are many experienced astrophotographers who uses the Seestar now as a grab-n-go alternative to their larger rigs. And there are beginners, who uses the Seestar as an entry to astrophotography. The Seestar in the hands of an experienced astrophotographer can create stunning images (although it is an EAA telescope). But a $5K rig in the hands of an newbie....
Great job Shawn. I had just updated GraXpert a few days ago to 3..0. I noticed that you were running 3.0.1 so I went out an upgraded and it was upgraded to 3.0.2. It shows that GraXpert is staying on top of the changes needed to make a better product. Congrats to GraXpert.
Thanks for the comparison. I have NoiseXterminator but I’m interested in giving Graxpert Denoise a try. It might suit some images more the Noise Xt so as it’s free, why not?
Hi Shawn, with your comparison you just show that NXT is more aggressive with the level of denoising in the default settings. And btw. nobody should use NXT with the default values because it too much denoising and details and tiny stars get lost. So it makes no sense to compare just the SNR with the default setting. I think a better approach would be to bring an image on the same SNR level and compare the details, stars and noise pattern. BR Thomas
Nice comparison Shawn. The SNR analysis certainly seems a fair way to compare apples with apples. When I tried similar comparisons, I found also there was a slight difference in texture left by the two programs. I feel like the texture with GraXpert felt a little more natural even where the noise reduction wasn’t quite so effective, so I am eager to see how GraXpert develops.
Clearly much better to have than not have GraXpert if you want a independent standalone program. Given the cost of Pixinsight why should you need so many third party applications?
Hi Shawn, I wonder whether you noticed any artefacts when using Graxpert Denoise? For me, when I compare the result to what I get with NXT, I almost always find many of them. To me it seems as if Graxpert Denoise takes random bits of noise and creates artefacts from them that look like faint fuzzies in the background. When I compare these results with photos from larger telescopes and longer integration times, these 'objects' just aren't there in reality. NXT doesn't create these artefacts at all. The artefacts show up no matter how high I set the strength value and also with the very newest AI model 2.0.0.
NoiseX for the clear win in this side my Side. Showing the before SNR of both and the after of both is a perfect way to grasp the validity of the results... Now with that being said? Graxpert is a GREAT option for a free use. No complaints from me when i use it. but i so see NoiseX inching it out in y own data.
Please don't use misleading profile photos on your video indicating that graxpert denoise literally doesn't work. It's literally an indirect ad to use noisexterminator instead.