Jump to content
Wumpus

MED resolution images auto-upsampled to HIGH; BW images import as colour

Recommended Posts

Affinity Photo Desktop Windows, latest (non-beta) rev:  1.6.5.135

Win 10 x64 up-to-date, 16 GB ram, >45 GB on SSD, >400 GB on HD & defragged no SMART errors). GTX 1060 6GB.

Camera: Sigma SD1M (M for Merrill, inventor of the Foveon sensor), X3F RAW files. Neither of these issues happen with my SD14.

Reproducible. Two unexpected behaviours both occurring during RAW import, even with the same image file.

Behaviour One

Import B&W RAW image, and B&W setting is lost (appears with separate colour channels). Camera has two methods for obtaining high quality B&W images:

1) At time of capture, choose B&W setting on camera body, which flags RAW metadata to treat R, G, B results as one value (grey) per photosite. Actual raw data would still have RGB per photosite, but image should only open as greyscale.

2) In Post-Processing, extract only the Blue channel from raw RGB data, from every photosite. This method loses the photons from the R&G layers, but since the Blue layer is the top layer, it has the least noise, so is pretty good. The other side-effect to this method is that because only the Blue layer is captured, it would be like using a deep blue filter over the lens, making yellows black, reds very dark, etc. So some things may look a bit different from a full-spectrum B&W image. The advantage is that it is reversible (you can decide you preferred the image as a colour image afterwards, whereas the first method keeps everything B&W for good).

I cannot find a way to achieve either method of getting a high quality B&W. I would prefer #1 (above) if I had to pick, as you get panchromatic response and lots of photons per photosite, resulting in ultra-low noise, just like from a Leica Monokrom. In this version, attempting this results in a lot of noise, even in a perfectly exposed image. In other RAW processors, I can use zero sharpening, zero de-noising (colour or luminance), and the images are clean. Also, default behaviour is to open as B&W, with only greyscale available (even Irfanview reads and presents the raw that way). I should note that it's not terrible noise, but the camera is capable of using these modes to deliver super creamy B&Ws, and I was hoping to be able to do that.

Behaviour Two

Import a RAW image taken at MED resolution, and the imported RAW will be auto-scaled up to FULL/HIGH resolution (I could not find how to defeat this default behaviour). Since it's not a perfect upscaling (I calculated around 47%), there are significant uprezzing artefacts, and it's not as sharp. The camera allows photos to be captured at LOW, MED or FULL/HIGH resolution.

Again, it's not horrible, but it's nowhere near how other RAW processors handle it. The camera uses one of the native features (though not all manufacturers make use of the 'feature') of CMOS sensors, 'binning', where arrays can report as if they were a lower resolution sensor: for example, a 4x4 grid of photosites would electronically be merged to act like a single photosite, so that all the photons get added together, giving you lower resolution but super low noise images and more accurate colour. Plus, Foveon images are often upscaled 200% after the fact, and still yield as good images as a Bayer CFA, so the resolution impact isn't near as negative as one might think. Foveons are the best at detecting subtle colours and recording micro-contrast without destructive processing, but sadly, they only do that when shot at low ISO or under loads of exposure, so these are frequent methods users employ.

Example: MED resolution photo (see green photo resolution) opened & upsampled/upscaled in Affinity Photo as a FULL resolution image (see yellow Affinity resolution). Also, this same photo was taken as a BW-only image, yet it is apparent that it contains hues.

Affinity-1.65_Upsampled+BW-to-Colour.png

One positive upside I discovered from the comparative testing is that the GPU acceleration in Affinity Photo is faster than the GPU-aided processing in both Corel and much moreso than even Sigma's own software. Haven't benchmarked Rawtherapee or On1 on that yet.

I have previously uploaded MED and FULL/HIGH resolution images, as well as colour and B&W images, from my camera to the Forum calling for sample images. Note that neither of these behaviours occur with my SD14 X3F (Foveon) images, so it is confined to the SD1M (aka Merrill). That exact same sensor is used in a few Sigma DP cameras as well (I believe it is the DP3 series), so they may exhibit identical behaviour for both these anomalies.

Thanks very much. Still learning a lot about the software, but it has made some things that were unbelievably tedious and error prone into downright easy when using Affinity Photo. Perhaps I have overlooked something and that is also the case here.

Share this post


Link to post
Share on other sites

Thanks @GabrielM

I was wondering the same last night, and read that the raw import engine was revamped, so I downloaded it, but hadn't installed it yet. I later read that Affinity installs it's betas alongside the production version, which is a great move, so no worries now about trying it. I will post back with an update.

Thanks very much again!

Share this post


Link to post
Share on other sites
Posted (edited)

 @GabrielM

Thanks again.

Update concerning the December 2018 1.7.0.x public beta (sorry, doing this from phone and don't have system available: happy to follow up with any missing information). Summary: no material change in these two behaviours, but some differences apparent.

Behaviour #1
The BW images still import as colour, and removing the blue-free channels (red + yellow) adds noise, and removing every non-blue channel (all channels to zero except for Blue) works similarly.

If I convert to BW (I think that was the 'tone' tab?), info tab then shows every pixel as having equal RGB values (e.g., R8 G8 B8). Interestingly, only the luminance noise removal has an effect (like a true BW image) and colour denoising does not affect it.

A quick test from an image tested earlier appears that denoising is much improved, but still requires a fair bit of noise removal levels (but about 20% less). I need to do more to get a more general and consistent assessment, but that's a side matter.

Behaviour #2

The MED Res raw images still open upsized to Full/High, but a difference is that before, the EXIF tab showed the 'real' (as captured in camera) resolution, and the raw data (as in the screenshot) showed the upsized resolution (conflicting) : now, both the EXIF and raw import data indicate the upsized (Full/High) resolution, so now they no longer conflict, but both are showing the maximum resolution, even though the images were captured at Medium resolution. If I use the Navigator to move down to the bottom right corner of a Medium resolution image, and put my cursor near that corner while the 'Info' tab is open, it shows the bottom right X , Y coordinates as matching the Full/High resolution, if that helps.

I didn't have time to open a Full/High resolution image to confirm if it still opens at the correct resolution.

High regards!

Edited by Wumpus
Added Info tab resolution observation in Behaviour #2

Share this post


Link to post
Share on other sites

I uploaded them to the "call for raw images" forum thread some weeks ago (not even knowing they would help with this; I just did it to ensure your team had some images from a couple different 1:1:1 Foveons).

I included BW and Medium resolution and full resolution images. I can send more later, but it will be several hours at best before I can.

Hope that helps a little for now. Thank you!

Share this post


Link to post
Share on other sites

I would need you to upload them on the link provided, as we do not have access to that "Call for RAW files" folder. 

Share this post


Link to post
Share on other sites

Thanks @GabrielM

 @Mark Ingram an Affinity Developer, has access to all those Dropbox files in his Call for RAWs post, so he may even have more samples than I had provided. I didn't bring the related images with me to work today, as I figured that if you're getting the sample images over to the Devs, and Devs already have the images, you might not need to go through the steps of getting them from me.

Call for Images forum post re: Sigma uploads

Since it sounds like we'll both need to jump through these officious hoops, I have hunted through all my local work computers, and found one image at Medium resolution, and it also happens to be BW, so maybe that's all you need. I did a test last night of the 1.7.0 beta, and confirmed that the Full/Max resolution raw images are not upsized in any way, so this behaviour appears to only affect the Medium resolution, and it appears that BW images are also brought in as RGB colour (same as happening with the production 1.6.5.x build). This sample is not in the best focus, and was a test before some other shots that night, but it meets the problem recreation criteria (BW, Med Res, Foveon Merrill sensor and related Affinity raw import processing). Social Media and File Sharing sites are blocked at my work, so DropBox isn't accessible from here, but I can send more tonight from home if necessary. Please do try to see if anything can be recreated in the interim, if you can. At least there's something attached.

Steps to recreate:

Open image in 1.6.5.x production or 1.7.0.x public beta.
Check EXIF and Affinity RAW dimensions (including 'info' tab). Should not be 4807x3206.
Check whether BW or Colour image. Should not be Colour.

For comparison, try another RAW processor, or even Irfanview.

Thanks

SDIM6720.X3F

Share this post


Link to post
Share on other sites

Perhaps another user might care to try to duplicate steps 1, 2 and 3? Happy to reciprocate with your error verification, if you're stuck too, as long as Windows 10 x64 and Affinity Photo 1.6.5x production are the requirements, and I can access an image in question.

We can do this!  :26_nerd:

Share this post


Link to post
Share on other sites

Oh, sorry for the delayed reply - I forgot to update the thread on Tuesday. I have already logged this with our developers. 

Share this post


Link to post
Share on other sites

Thanks so much @GabrielM
 Wasn't sure if everyone was occupied prepping for the 1.7.0x launches. Really appreciated. Will take and post further samples this weekend to see if the BW issue doesn't affect High or Low resolutions: perhaps it's limited to Med resolution.

Thanks again!

 

Share this post


Link to post
Share on other sites

Update: Took two new photos, both BW RAW using in-camera settings. One was LOW resolution, one was FULL/HIGH resolution.

Software tested was both Affinity Photo beta 1.7.0.258 and Affinity Photo 1.6.5.135. Just for the record, images were shot using final published firmware (1.09.0.7652).

First , the common findings for LOW and HIGH Resolution (mind you, just from one sample image):

BW images still import as RGB colour (I also found that Kalpanika X3F extractor does also, so Affinity Photo is no longer alone in that).  Affinity Photo is still noisier than all other tested raw developers (even Kalpanika) when they have both denoising and any spatial gain disabled. Affinity requires high levels of denoising to provide images with similar levels of noise to other raw processors.

Both LOW & HIGH resolution images correctly show their resolutions (in EXIF, and Affinity Photo image info). There was no up-rezzing for either., like we observed with MED resolution images. And separately, I think I found the root cause.

Further investigations show that Sigma uses a slightly different method of binning for only the MED resolution images. FULL/HIGH resolution images are not binned (normal behaviour, as they are using all sensor sites), and they work in both Affinity Photo Windows app versions as expected. LOW resolution images are actually captured at high resolution in the camera, but a 2x2 sensor site array merges reported photons for all four photosites to report as low resolution to the in-camera processor, and the output file also then acts as if it is a low resolution image (rather than just interpolating 100% like Bayer cameras). This gives you four times the sensor site area and improved pixel signal/noise ratio, but the image is a quarter of the resolution (containing truer colour  than downsampling in post process). I couldn't even find the published array dimensions for the MED resolution binning, but a contact suggested it was an unusual ratio something like 3 horizontal on the bottom row and two horizontal on the row immediately above, or something non-standard like that. I honestly don't know (I wish I did) exactly how MED resolution is binned. Doing some raw 'raw' dumps without any corrections in other tools yields images taken at MED resolution output as roughly double-wide, and would require some degree of vertical interpolation to deliver the standard published 'MED' output resolution dimensions.

I think this non-standard interpolation (not equally in both dimensions) is what causes Affinity Photo to fall back to uprezzing the entire image to FULL/HIGH resolution. It actually does surprisingly well at that, considering, but it isn't quite the same quality as other tools. However, I can appreciate how the raw extraction would be unusually complex, and it's nice to see Affinity not using everyone else's raw engine, so I understandably won't expect further efforts from Affinity to be spent on that issue. I will switch to just using LOW or FULL/HIGH resolution, and that's OK. It might be good to put a note in your support documents about that in case you get future calls.

I would still like to see if Affinity can improve the BW import, and now it can focus on just LOW & HIGH resolution. The images are still very noisy, and there are two potential methods to get low-noise BW images. If we can even prioritize the first method (as the primary mode for BW capture designed by the manufacturer) that would let users get panchromatic and very low noise BW images, and since many others do it, hopefully it's not computationally prohibitive. 

Thanks for your perseverance on this matter. Please let me know how Affinity plans to proceed on this case. Again, this will almost certainly affect all other Sigma cameras using the Merrill sensors, so is not quite as niche an occurrence as it might first seem.

Best Regards

Share this post


Link to post
Share on other sites

Just a little thread bump in my quest for a response. Looking around the web, apparently the raw import quality from Foveons has been an ongoing issue for some time. How can we work together to improve it? Happy to contribute efforts too.

Share this post


Link to post
Share on other sites
On 3/12/2019 at 6:32 AM, Wumpus said:

....(edited)....

I would still like to see if Affinity can improve the BW import, and now it can focus on just LOW & HIGH resolution. The images are still very noisy, and there are two potential methods to get low-noise BW images. If we can even prioritize the first method (as the primary mode for BW capture designed by the manufacturer) that would let users get panchromatic and very low noise BW images, and since many others do it, hopefully it's not computationally prohibitive. 

Thanks for your perseverance on this matter. Please let me know how Affinity plans to proceed on this case. Again, this will almost certainly affect all other Sigma cameras using the Merrill sensors, so is not quite as niche an occurrence as it might first seem.

Best Regards

Just confirming that the current status of this issue is still being investigated, and that (only) issue now is obtaining BW images from Merrill sensors, at Low or High/Full resolution. Can't do it yet (other than using kluges that result in really noisy images). The X3F Merrill images are capable of native BW at really high quality.

Thank you.

 

Share this post


Link to post
Share on other sites

Okay, thanks @GabrielM, I will go forward on the assumption that they received the newer info (updated after the first time you forwarded it to them), and that they are only focusing on the first half of the original post's two issues (inability to get BW, not the upsizing issue).

Appreciated.

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now

×

Important Information

These are the Terms of Use you will be asked to agree to if you join the forum. | Privacy Policy | Guidelines | We have placed cookies on your device to help make this website better. You can adjust your cookie settings, otherwise we'll assume you're okay to continue.