NotMyFault Posted November 7, 2023 Posted November 7, 2023 Hi, it seems the live noise reduction filter has no effect. steps to reproduce create new file UHD size create rectangle in canvas size add gradient fill black to white horizontal add live add noise filter, set to 100% merge visible now add destructive or live noise removal filter, set luminance to 100% expected: both should remove noise to large extend. observed: live filter shows no effect. V2.2 RPReplay_Final1699378337.mov loukash 1 Quote Mac mini M1 A2348 | MBP M3 Windows 11 - AMD Ryzen 9 5900x - 32 GB RAM - Nvidia GTX 1080 LG34WK950U-W, calibrated to DCI-P3 with LG Calibration Studio / Spider 5 | Dell 27“ 4K iPad Air Gen 5 (2022) A2589 Special interest into procedural texture filter, edit alpha channel, RGB/16 and RGB/32 color formats, stacking, finding root causes for misbehaving files, finding creative solutions for unsolvable tasks, finding bugs in Apps. I use iPad screenshots and videos even in the Desktop section of the forum when I expect no relevant difference.
loukash Posted November 7, 2023 Posted November 7, 2023 Same issue as on desktop. NotMyFault 1 Quote MacBookAir 15": MacOS Sonoma > Affinity v1, v2, v2 beta // MacBookPro 15" mid-2012: MacOS El Capitan > Affinity v1 / MacOS Catalina > Affinity v1, v2, v2 beta // iPad 8th: iPadOS 18 > Affinity v2
Dan C Posted November 8, 2023 Posted November 8, 2023 Thanks for your report @NotMyFault! 18 hours ago, NotMyFault said: expected: both should remove noise to large extend. observed: live filter shows no effect. As I understand it, this is expected behaviour which has been covered by our developers previously here - On 6/18/2019 at 1:06 PM, Justin said: This is to be expected. The destructive filter estimates the noise level throughout the image and scales the intensity of the effect accordingly. The live filter cannot do this as the full image is not available to it due to the tiled nature of the rendering pipeline. I hope this clears things up Quote
loukash Posted November 8, 2023 Posted November 8, 2023 1 hour ago, Dan C said: this is expected behaviour which has been covered by our developers previously here Alright, but it still should do at least something, right? From my testing here, it doesn't do anything at all, at any setting: Paul Mudditt 1 Quote MacBookAir 15": MacOS Sonoma > Affinity v1, v2, v2 beta // MacBookPro 15" mid-2012: MacOS El Capitan > Affinity v1 / MacOS Catalina > Affinity v1, v2, v2 beta // iPad 8th: iPadOS 18 > Affinity v2
Dan C Posted November 8, 2023 Posted November 8, 2023 9 minutes ago, loukash said: Alright, but it still should do at least something, right? From my testing here, it doesn't do anything at all, at any setting: The Live Denoise filter is working in my testing - though as above you will see different results when compared to the destructive version. Live Denoise not applied (100% Zoom) - Live Denoise applied (100% Zoom) - Both examples above Rasterised, with top layer set to 'Difference' and then inverted / zoomed (as the noise difference is harder to spot at lower zoom levels on a black background) - (400% Zoom) - The above shows that the Live Denoise Filter is correctly affecting the image in my example. I've tested this across both macOS and Windows, using the below file - Live Denoise.afphoto You may need to adjust your Live Denoise filter settings to see changes apply to the image, as the Live filter does not scale the effect of the Denoise, unlike the destructive version, as mentioned above. loukash 1 Quote
loukash Posted November 8, 2023 Posted November 8, 2023 6 minutes ago, Dan C said: You may need to adjust your Live Denoise filter settings to see changes apply to the image I see: in my example, I should have set Luminance Detail to 0% to see at least some effect. It's still very faint though, and only affects the parts where neighbor pixels have low contrast. So for such extremely noisy images, it needs e.g. the Dust & Scratches live filter first, then smoothing with Live Denoise on top. Quote MacBookAir 15": MacOS Sonoma > Affinity v1, v2, v2 beta // MacBookPro 15" mid-2012: MacOS El Capitan > Affinity v1 / MacOS Catalina > Affinity v1, v2, v2 beta // iPad 8th: iPadOS 18 > Affinity v2
loukash Posted November 8, 2023 Posted November 8, 2023 However… General confusion will still remain as long as the filters and their values are named the same. Paul Mudditt 1 Quote MacBookAir 15": MacOS Sonoma > Affinity v1, v2, v2 beta // MacBookPro 15" mid-2012: MacOS El Capitan > Affinity v1 / MacOS Catalina > Affinity v1, v2, v2 beta // iPad 8th: iPadOS 18 > Affinity v2
NotMyFault Posted November 8, 2023 Author Posted November 8, 2023 In many images the live Filter shows next to none effect, especially when crafting test document by simply using add noise filter on a smooth surface. We have other real-live where noise removal failed totally (will search Forum and add links later) in my example image of this thread, live noise removal had absolutely no effect at 100%, even with sliders to extreme (100% for strength, 0% for details), whereas non-live removed all noise with ease, at 13% the noise is almost eliminated. 3 hours ago, Dan C said: I hope this clears things up Sorry to say, but not this time. denoise bug.afphoto Quote Mac mini M1 A2348 | MBP M3 Windows 11 - AMD Ryzen 9 5900x - 32 GB RAM - Nvidia GTX 1080 LG34WK950U-W, calibrated to DCI-P3 with LG Calibration Studio / Spider 5 | Dell 27“ 4K iPad Air Gen 5 (2022) A2589 Special interest into procedural texture filter, edit alpha channel, RGB/16 and RGB/32 color formats, stacking, finding root causes for misbehaving files, finding creative solutions for unsolvable tasks, finding bugs in Apps. I use iPad screenshots and videos even in the Desktop section of the forum when I expect no relevant difference.
NotMyFault Posted November 8, 2023 Author Posted November 8, 2023 (edited) And please note that I report for iPad specifically (M1 model A2348). i will test on Mac later. Edit: Mac shows identical results. Edited November 8, 2023 by NotMyFault Paul Mudditt 1 Quote Mac mini M1 A2348 | MBP M3 Windows 11 - AMD Ryzen 9 5900x - 32 GB RAM - Nvidia GTX 1080 LG34WK950U-W, calibrated to DCI-P3 with LG Calibration Studio / Spider 5 | Dell 27“ 4K iPad Air Gen 5 (2022) A2589 Special interest into procedural texture filter, edit alpha channel, RGB/16 and RGB/32 color formats, stacking, finding root causes for misbehaving files, finding creative solutions for unsolvable tasks, finding bugs in Apps. I use iPad screenshots and videos even in the Desktop section of the forum when I expect no relevant difference.
NotMyFault Posted November 8, 2023 Author Posted November 8, 2023 3 hours ago, Dan C said: Thanks for your report @NotMyFault! As I understand it, this is expected behaviour which has been covered by our developers previously here - I hope this clears things up searched and found again the older report from another user from March 2023. At that time you rated this as something to be investigated: Quote Mac mini M1 A2348 | MBP M3 Windows 11 - AMD Ryzen 9 5900x - 32 GB RAM - Nvidia GTX 1080 LG34WK950U-W, calibrated to DCI-P3 with LG Calibration Studio / Spider 5 | Dell 27“ 4K iPad Air Gen 5 (2022) A2589 Special interest into procedural texture filter, edit alpha channel, RGB/16 and RGB/32 color formats, stacking, finding root causes for misbehaving files, finding creative solutions for unsolvable tasks, finding bugs in Apps. I use iPad screenshots and videos even in the Desktop section of the forum when I expect no relevant difference.
Dan C Posted November 8, 2023 Posted November 8, 2023 1 hour ago, NotMyFault said: in my example image of this thread, live noise removal had absolutely no effect at 100%, even with sliders to extreme (100% for strength, 0% for details) Using the document you've provided (where Luminance is 100% and Luminance Detail is 50%), applying the same testing methodology in my previous post, on my iPad here, I see the following: Live Denoise not applied (100% Zoom) - Live Denoise applied (100% Zoom) - Both examples above Rasterised, with top layer set to 'Difference' and then inverted / zoomed (as the noise difference is harder to spot at lower zoom levels on a black background) - (400% Zoom) - As shown in the inverted 'Difference' results, the filter is affecting the document, but not to the same level as the destructive filter (or perhaps meeting your expectations for the live filter). 1 hour ago, NotMyFault said: whereas non-live removed all noise with ease, at 13% the noise is almost eliminated. Again, this is due to the destructive version of the filter being scaled, based on the layers contents, unlike the Live Filter. For examples/images such as the above, I'd recommend using the destructive version of the filter when you need to remove extreme levels of noise. 48 minutes ago, NotMyFault said: searched and found again the older report from another user from March 2023. At that time you rated this as something to be investigated: This quoted thread was a report regarding the difference between the Live Filter in v1.6.5 compared to the Live Filter in v2, which is still open with our development team. I believe the 'legacy' version of the Denoise Live Filter, which offered an 'extreme' checkbox, for cases such as these, uses a different algorithm for noise reduction compared to the current Live Filter. I can't provide any more information regarding the legacy Live Filter at this time, as this is all the information I have been provided from our developers. If you believe the Live version of the Denoise Filter should have this 'extreme' option returned, then I'd recommend creating a feedback post regarding this - as our developers don't class this change from the legacy live filter to the new version of the live filter as a bug. Quote
NotMyFault Posted November 8, 2023 Author Posted November 8, 2023 Thanks for following up. I’m still not convinced by the arguments and examples you provided. The effect of live filter is barley measurable, by factor 50 less visible than the non-live filter. The live filter is effectively unusable. For me it is not relevant that historical versions offer better results. I don‘t want an „extreme“ setting. i simply expect minimal consistency: if live and non-live filter share identical parameters and name, the effect should be identical a live filter giving 1/50th of strength is useless. If Affinity insist of not seeing any bug and denies any need to correct this on your own, go ahead, I have to accept and will remain silent in future. In this specific case i will not file a feature request from my side, it would be too far beyond reasonable user expectations, UI consistency, and logic. Just leaning back and watching new cases of others falling into the same trap coming in from time to time. Paul Mudditt 1 Quote Mac mini M1 A2348 | MBP M3 Windows 11 - AMD Ryzen 9 5900x - 32 GB RAM - Nvidia GTX 1080 LG34WK950U-W, calibrated to DCI-P3 with LG Calibration Studio / Spider 5 | Dell 27“ 4K iPad Air Gen 5 (2022) A2589 Special interest into procedural texture filter, edit alpha channel, RGB/16 and RGB/32 color formats, stacking, finding root causes for misbehaving files, finding creative solutions for unsolvable tasks, finding bugs in Apps. I use iPad screenshots and videos even in the Desktop section of the forum when I expect no relevant difference.
Dan C Posted November 8, 2023 Posted November 8, 2023 47 minutes ago, NotMyFault said: The effect of live filter is barley measurable, by factor 50 less visible than the non-live filter. I can only provide the information that our developers supply - and as previously quoted above there is an expected difference between the Live and Destructive versions of this filter. 48 minutes ago, NotMyFault said: if live and non-live filter share identical parameters and name, the effect should be identical I understand these expectations, and I will share this feedback with our team to suggest that we might perhaps rename the Live version of the Filter to provide some distinction to users. 49 minutes ago, NotMyFault said: a live filter giving 1/50th of strength is useless. As confirmed previously, this is already logged (and remains 'open') as a bug with our developers - when comparing the strength of the Live Filter compared to the previous Live Filter offered in 1.6.5. Simply the issue cannot be logged as a bug when comparing the Live and Destructive versions of the filter, due to the expected differences outlined above. Quote
NotMyFault Posted November 8, 2023 Author Posted November 8, 2023 Thank you Dan. I really appreciate your polite and absolutely professional attitude to explain everything and provide as much transparency as possible about those Serif internals (even if the information leaves me speechless in rare cases). Have a great evening. Dan C 1 Quote Mac mini M1 A2348 | MBP M3 Windows 11 - AMD Ryzen 9 5900x - 32 GB RAM - Nvidia GTX 1080 LG34WK950U-W, calibrated to DCI-P3 with LG Calibration Studio / Spider 5 | Dell 27“ 4K iPad Air Gen 5 (2022) A2589 Special interest into procedural texture filter, edit alpha channel, RGB/16 and RGB/32 color formats, stacking, finding root causes for misbehaving files, finding creative solutions for unsolvable tasks, finding bugs in Apps. I use iPad screenshots and videos even in the Desktop section of the forum when I expect no relevant difference.
loukash Posted November 8, 2023 Posted November 8, 2023 1 hour ago, NotMyFault said: if live and non-live filter share identical parameters and name, the effect should be identical Agreed. And if they cannot be identical for technical reasons, then one of those should be renamed correspondingly. NotMyFault 1 Quote MacBookAir 15": MacOS Sonoma > Affinity v1, v2, v2 beta // MacBookPro 15" mid-2012: MacOS El Capitan > Affinity v1 / MacOS Catalina > Affinity v1, v2, v2 beta // iPad 8th: iPadOS 18 > Affinity v2
Dan C Posted November 9, 2023 Posted November 9, 2023 Thanks for your kind words NotMyFault, it's certainly appreciated 15 hours ago, loukash said: And if they cannot be identical for technical reasons, then one of those should be renamed correspondingly. Just to confirm, I've logged this as a requested improvement with our development team internally & I'll be sure to add your 'vote' to this now! NotMyFault and loukash 1 1 Quote
lepr Posted November 11, 2023 Posted November 11, 2023 On 11/8/2023 at 6:39 PM, loukash said: one of those should be renamed correspondingly. A suggestion: The Useless Live Filter loukash and NotMyFault 2 Quote
Recommended Posts
Join the conversation
You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.