 Info It's a stitching of 3 vertical (hand-held) images (I've added background information and Pano template to the image description) to increase detail sharpness and to avoid border unsharpness. I've used Hugin for stitching and as far as I can remember Hugin moderately downscales the output image which is imho eligiable to create crisp images with good detail sharpness and no problem due to the high resolution. I made some final, very subtle local (on the church building) brightness adjustments with Viveza from NIK filters. --Tuxyso (talk) 21:33, 20 August 2014 (UTC)[reply]
- @Tuxyso: Hugin does nothing you doesn`t tell it before. You can scale the output image just the same way like in PTGui for example. So if Hugin downscales your pictures then something's "wrong" with your prefs. My tip: try to stitch the pano with the highest possible settings (compression, resolution...) and if downsampling is necessary do it in the final past processing. regards mathias K 06:09, 21 August 2014 (UTC).[reply]
- Sorry mathias but you are wrong. Hugin downscales the output image to 70% (per default!). You can surely change it, but if you take a look e.g. at this newsgroup post there seem to be good reasons for a slight downscaling regarding lens/sensor limitations. You never archive the full sensor resolution (despite you own a Zeiss Otus lens) - take a look on DxOMark and compare different lenses. Thus it is no bad idea to keep the default setting of Hugin as it is. IMHO photos have to look "crisp" / sharp at 100%. Nearly every pano here on Commons is slightly downscaled and exactly that is the reason why panos often have brilliant detail sharpness. 13,9 Mpx are definitely enough. Your suggestion to do slight downscaling in past-processing (and not by Hugin) is worth a try - thanks for the hint - I will make some comparisation in the future. --Tuxyso (talk) 06:25, 21 August 2014 (UTC)[reply]
- I see that newsgroup posts the old myth about the Bayer filter justifying a 70% downsample. Well a 70% downsample of a 36MP image is 17.7MP. Yet somehow the Sony FE 55mm or the Zeiss Otus 55mm both achieve a 29MP rating on a 36MP camera according to DOXMark. And a Sigma Art 50mm achieves 17MP rating on a 21MP Canon (which is only 10MP after 70% reduction). Lens quality, diffraction, sensor noise, anti-aliasing filter, software lens-correction, perspective correction, rotation, and JPG compression all result in different forms of image degredation. None of these lose information in quite the same way as a downsample/upscale process would. Lens resolution varies across the image (centre sharper) whereas downsampling loses resolution equally. The desire for perfection at 100% is just an artefact of how we look at them (click on the image and bang! 100%) and the culture of pixel peeping. An ideal situtation would be for the uploader to be able to set an best-viewing-resolution for any given image and the software defaults to zooming at that level. This would allow Tuxyso to upload a 20MP image and have the Wiki software present it at 14MP (with a touch of output-sharpening). This downsample-and-sharpen-for-presentation is how most photo websites work, including MediaWiki. Having said that, the biggest degredation with a stitched image is usually the perspective correction and rotations that can occur, which can in some (but not all) images massively stretch parts. While there's no point uploading a very soft/noisy image, I don't think asking for all our featured photos to be this crisp at 100% is a desirable goal. Please nobody think that downsizing 70% is necessary for FP. It is a 50% reduction in MP. I'd have had to reduce my A33 images to 7MP if that was necessary and I've got plenty single-shot images that are sharp at 14MP without resorting to Zeiss glass. Tuxyso, for modest stiches like above, I suggest you try a much more conservative downsample at Lightroom-export stage. -- Colin (talk) 12:22, 21 August 2014 (UTC)[reply]
- I often save panorama with no downsampling in Hugin and get good results. Here is a 81mp panorama created from 5 hand held vertical photos.--ArildV (talk) 15:10, 21 August 2014 (UTC)[reply]
- I use PTGui for my panoramas, which is very similar to Hugin in its options for generating the final output. Here the default is "optimal", which effectively downsamples to the level of pixel crispness shown here. I always change it to "maximum", which I think correspond to not downsampling in any areas of the blended photograph and then in the final stages of the processing (GIMP in my case) after rotating, tweaking perspective I have a look at the pixel quality at full resolution, and then try-out different levels of downsampling, if any. If I have used my 18-55 mm kit lens, which is honestly speaking a pretty bad lens when it comes to CA and other lens distortions, I sometimes go down to half the original number of pixels in the worst cases (example), whereas if the images are based on my better 55-250 mm lens I downsample much less, typically down to 75% of the original pixel count depending on field-of-view and ISO (example). That does not give me the same per pixel crispness as in the candidate here, but at the same time I believe the information loss is marginal, meaning it will be optimal for large scale printing (in the unlikely event, anyone wants to do that, lol) avoiding visible pixelation. I agree with Colin that there should be better options for reviewing at a fixed resolution, to discourage the pixel peeping. I once had a 35Mpixel nomination, not downsampled, with rather bad per-pixel quality, which I 'refused' to downsample, because information is lost. There was a long-winded discussion with many examples given for what happens with down- and upsampling, which might have your interest. --Slaunger (talk) 15:33, 21 August 2014 (UTC)[reply]
- Ha ha! I argued the opposite way three years ago! I agree with my old self that there does come a point when the 100% image is mush and we're kidding ourselves that it is worth keeping/uploading at that level. But with a good camera and a good prime lens, many people get excellent results with no downsizing at all. Slaunger, when you say "half the original number of pixels" I take it you mean a 50% megapixel reduction (i.e. about 70% reduction in each linear dimension). I thought Tuxyso had reasonable kit and so was surprised he felt it necessary to half his megapixels before daring to show his work here ;-) -- Colin (talk) 18:49, 21 August 2014 (UTC)[reply]
- Colin, actually I think you have been pretty consistent:) My nom was with a crap 2007 compact camera sensor. I did mean 50% reduction in the number of pixels, i.e. approxiately 71% the number of original pixels on each side. Yes, I do think that with Tuxyso's gear, there is no need to downsample so much. Information is lost irreversibly in the process in getting all the crispiness. --Slaunger (talk) 19:04, 21 August 2014 (UTC)[reply]
- Info Just for information purposes I've uploaded a full res version. It just confirmed my standard procedure :) I am not convinced with the detail sharpness there. If I click on photos on Commons they should look crisp even at 100% view (I know this point is controversal). Probably I will set Hugin to less downscaling when I take photos with the exceptional Sigma 18-35 f1.8 lens which I will receive today :))) I also compared downscaling in Hugin with Hugin fullres and output downscaling with Lightroom. I could not see any significant differences (despite the larger file size). --Tuxyso (talk) 08:16, 22 August 2014 (UTC)[reply]
- Have fun with the 18-35! I guess you're the first one here to use one, but what I've seen so far is pretty nice! Imho the full-res image is still good and downsampling doesn't add value. --DXR (talk) 18:08, 22 August 2014 (UTC)[reply]
- Wow! The full res is very good. No need to downsample such a beauty:) --Slaunger (talk) 20:19, 22 August 2014 (UTC)[reply]
|
Support -- Tuxyso (talk) 21:33, 19 August 2014 (UTC)
ArionEstar (talk) 00:34, 20 August 2014 (UTC)
