Can you guys change the “noise reduction” processing back to the way it worked at version 3.0.4? The results used to be much better. All it’s doing lately is just CREATING noise where there wasn’t any before. Instead of reducing noise, the results are actually deblurring and edge sharpening. Bug?? If this is a feature, I hate it. The photo is noise reduction ONLY (Proteus/Manual)…creating this weird noise pattern and edge sharpening for no reason.
I just spent the last 2 hours going back through the versions. This behavior started at version 3.0.5 and has only gotten slowly worse which is why I never noticed until now.
Version 3.0.4 was the last one with actual denoise results (upscaling SD to HD) that didn’t create noise.
Gaia 1x reports ERR fps…what would be causing this?
Edit: ran the benchmark again and no ERR this time. I probably ran into an edge case scenario where the numbers/results/significant digits did not fit the formatted output field. You guys might want to look into that.
Is it expected behaviour, that when adding noise (for example, a value of 2), the noise is actually visible for the first few frames? This is using Artemis on an Intel Mac with AMD GPU. It doesn’t seem right.
Good speed overall on this release, good work and thanks for incorporating the speed boost from the Alpha release. Now, on my wish list for the v3.3.0 release:
Scene change detection for Chronos (all variants) so we don’t get blurry interpolated frames at scene change points
Frame numbering for navigation/scrubbing/editing
Updates to any and all models
Edit: and of course the improved preview updates that are already in the pipeline!
Thanks and keep up the good work. Rome is taking shape nicely!
On my 3060 ti system I see about 50% speed improvement in Artemis High at 1080p, though I am mainly interested in upscaling and sometimes slowing down older videos sometimes as low as 288p. With them, there is a small 5% improvement, enough to take me up over the 100 fps mark at 2x upscale - handy!
On slomo, I feel that the benchmark numbers are unhelpful, and for some may be misleading. Yes there is a small speed increase on my system, again about 5%, but what really matters I think is the fps comparison based on the original video frame numbers (assuming that the same fps rate will be maintained after processing).
So based on that, at 576p, Chronos fast is now about 2 to 3 times faster than Apollo at 2x slomo, is about the same at 4x slomo, and is around half as fast at 8x slomo. The real life slomo speed range on my system ranges from 13.4 original frames a second on 4x Apollo (and about the same on all Apollo speeds) to 40 original fps on 2x Chronos fast. That makes it easy to work out how long a conversion will take - and which model/speed to use given that many of us have limited time to work on these things.
Forgot to mention, for anyone holding back on upgrading to v 3.x, Artemis High at 288p and 2x upscale now processes 3.3 times quicker than version 2.6.4.
These are not from the benchmarks but are my actual processing times of 94fps vs 28 fps, at 288p/2x upscale. 3060 ti GPU, which maxes out now (and CPU R7 5800x at 33%).
Here’s to hoping my RX 6900 XT doesn’t have it’s black section issue anymore with this version.
The benchmark says the speed is doubled. That’s a pretty sweet improvement for AMD cards!