Glen Barrington
Forum Pro
No history from my side that I'm aware of. The written word is, at best, a poor approximation of verbal communication. This whole thread was meant to be fun. not start 'stuff'.
-
Welcome to the new forums! Please read this first. For known issues we are working to resolve, click here.
Surprising Resize comparison. See photos below.
- Thread starter Glen Barrington
- Start date
Glen Barrington
Forum Pro
Quite possibly, but I have Photo AI and don't have Gigapixel. I'm looking for ease of use in a casual situation. Gigapixel sounds like something for people with ongoing needs. I just don't care THAT much.
I could have, but that would require work, I was trying to recreate how I thought a casual user would approach this task.
This was supposed to be a fun little thing. I don't feel the need to defend my position. Nor do I feel the need to share my source images. I assume you guys take photos of all sorts, prove this stuff based on YOUR needs.
BTW, argue amongst yourselves. I'm done responding.
--
Life is an ongoing learning process. You can stop once you are dead.
http://glenbarrington.blogspot.com/
http://glenbarringtonphotos.blogspot.com/
https://www.flickr.com/photos/130525321@N05/
Last edited:
Digital Nigel
Forum Pro
That’s not a stiff enough test to bother with. Most competent algorithms produce reasonable results with such a small magnification. For a serious evaluation, you need to set a tougher test, even if you wouldn’t often use 3x.
Why Photo AI, when Gigapixel AI usually produces superior results? It’s generally regarded as the industry standard, particularly for high magnifications. Even at 4x it produces surprisingly good results.
Yes, with any images intended for web posting or email.
It’s not the default or a recommendation, just the first in an alphabetical list of standard mathematical resampling algorithms. In effect, you were testing algorithms, not Affinity the product. And you then picked the worst option available. Why bother downloading and installing a product, and then deliberately use it in the worst possible way in a test? Of course it will come last.
This is Google’s AI summary:
Bilinear, bicubic, and Lanczos are image interpolation algorithms used for resizing images, with Lanczos generally offering the highest quality but at a higher computational cost, while bicubic provides a good balance between quality and speed, and bilinear is the simplest and fastest but with the least quality.
Here's a more detailed comparison:
- Bilinear Interpolation:
- How it works: Uses a 2x2 neighborhood of pixels to calculate the new pixel value by averaging the surrounding pixels.
- Pros: Fast and computationally inexpensive.
- Cons: Can result in some softening of details and may introduce some jagged edges.
- Bicubic Interpolation:
- How it works: Uses a 4x4 neighborhood of pixels and applies a more complex weighting to calculate the new pixel value, resulting in smoother transitions.
- Pros: Offers a good balance between quality and speed, and is often the default algorithm for many image processing tools.
- Cons: Can still introduce some softening, especially when downscaling significantly.
- Lanczos Interpolation:
- How it works: Uses a larger neighborhood of pixels (e.g., 8x8 or 32x32) and applies a more sophisticated weighting function to calculate the new pixel value, resulting in very high-quality results.
- Pros: Generally produces the best quality results, especially for upscaling or when significant scaling is involved.
- Cons: Computationally expensive, and can be slower than bicubic or bilinear.
- Choose Bilinear for: situations where speed is paramount and quality is not a primary concern.
- Choose Bicubic for: situations where a good balance between quality and speed is needed, and it's often the default choice for many applications.
- Choose Lanczos for: situations where the highest possible quality is required, even if it means a slight performance hit.
A casual user would miss out on potentially much better results obtainable by taking advantage of the options available in the tools being used.
Digital Nigel
Forum Pro
Very much so, plus Affinity Photo isn’t really a tool for casual users. You certainly wouldn’t buy it as a resizing tool; that’s just one of the many standard capabilities expected in any photo editor. Affinity offers several standard, well-known mathematical algorithms, just like most photo editors, but doesn’t have any unique, proprietary options of its own.
AlGears
Active member
- Messages
- 80
- Reaction score
- 39
Anybody else here see this thread ending this way from the beginning?
Breakfastographer
Leading Member
- Messages
- 689
- Solutions
- 1
- Reaction score
- 413
No comment from me except to say, if you're working with RAW rather than JPEGs, your choice of converter will majorly affect your starting point and hence expected results - I collected some data on this six years ago and do not intend to repeat the exercise any time soon since it was a major effort:
breakfastographer.wordpress.com
35 RAW converters reviewed, part 2: fine detail
Part 2 is about the fine detail. The following are 100% close-ups of the images in part 1, most important ones presented first as a group. If you are reading this without having read part 1, you…
breakfastographer.wordpress.com
Digital Nigel
Forum Pro
That was a lot of work!
Of course, such analyses soon get dated, as some products disappear, others get significantly updated, and new ones appear. For example, Topaz Gigapixel is probably the best upscaler today, but I don't think it even existed when you did your analysis.
But I was amused at your findings re the Skylum products, which seem to be as accurate today as when you first made them.
tbcass
Forum Pro
I appreciate the effort you put into that but, because the most recent was nearly 5 years ago, it isn't relevant to today's RAW software. There has been considerable advancement in RAW software since then.
--
Tom
Last edited:
tbcass
Forum Pro
You won't find casual users in this forum! 
Breakfastographer
Leading Member
- Messages
- 689
- Solutions
- 1
- Reaction score
- 413
I don't know why that comment was needed - I already stated I wouldn't repeat the exercise, what more is there to say?
grsnovi
Veteran Member
I'm not so sure.
Affinity Photo is often a choice made by casual users who don't want to subscribe to Adobe Photoshop based solely on their often sale price which is quite attractive.
I'm probably not a casual user despite the fact that I don't use 90% of what my tools can do.
Until I read Digital Nigel's thread below, I had no idea what or why the various algorithms affected the up and/or down-sizing of files. 99.9% of the time I only ever used Affinity to downsize a camera's native size to something small to share online and to be honest, the bilinear option always looked just fine.
Even now with a faster machine and additional s/w for doing (predominantly) noise control during raw conversion I still don't think of myself as a sophisticated user but I'd like to think I'm somewhat savvier than a retouching novice.
For several year's (while employed in my last job) I had Adobe's CC Suite. I used PS a little preferring the simplicity of Paintshop Pro. I used Illustrator a lot (several times a week) and Premier several times a year.
Sometimes it's hard to imagine that those of you who are obviously experts on the current crop of commercially available (and open-source) photo-centric s/w do that much photography.
In any event, I'm grateful that those of you who are much more expert than I am are willing to take the time to help me when I get lost.
Would you care for to check results of the other algorithms for Affinity and give us your updated thoughts?
It is reasonable to expect that someone who wants to resize a photo to make a big impact, e.g. to print it in 40-60in. format for $$$, would recheck what is going before pressing 'print' or 'send' button. I guess that is how it works in the real world.
As it is now, your test and conclusions regarding Affinity are not convincing for me.
It is reasonable to expect that someone who wants to resize a photo to make a big impact, e.g. to print it in 40-60in. format for $$$, would recheck what is going before pressing 'print' or 'send' button. I guess that is how it works in the real world.
As it is now, your test and conclusions regarding Affinity are not convincing for me.
Digital Nigel
Forum Pro
Those resampling methods are not specific to Affinity Photo. They are well-known mathematical formulae that have been analysed in great detail in the textbooks. There would be no reason for me to bother repeating those tests.
What test are you referring to? I didn’t publish any test.
In any case, I don’t use Affinity Photo for upsizing. I use either Photo AI or Gigapixel AI.
In the past, when I’ve done tests, I found that it didn’t make very much difference which tool or method you used up to about 2x. But above that, the Topaz AI upscalers were much superior, and continue to improve.
Jacques Cornell
Forum Pro
I provide my event clients with "print-resolution" (6000px) and "web-resolution" (2048px) JPEGs, both of which are downsized from my 33MP and 61MP RAW files. The print-rez files save storage space and download time for the client while remaining big enough for almost any expected use. The web-rez files appear crisper online for two reasons: 1) the web host's engine doesn't have to resize them, and 2) I apply smart output sharpening that takes pixel dimensions into account on export from Lightroom Classic, using the "screen" setting.
[trimmed]
--
Event professional for 20+ years, travel & landscape enthusiast for 30+, stills-only.
http://jacquescornell.photography
http://happening.photos
Last edited:
tbcass
Forum Pro
The point was why did you bother posting it in the first place. People who don't know better might mistakenly think the comparison is relevant today. My response was meant to inform those people to ignore the results.
Help within Affinity Photo says this:
The following resample settings are available:
- Nearest Neighbor—simple resampling which has the fastest processing time. Use for hard-edge images and pixel work.
- Bilinear—algorithmic resampling for use when downsampling images.
- Bicubic—algorithmic resampling for use when upsampling images. Resampling is smoother than Bilinear but has a slower processing time.
- Lanczos 3—complex algorithmic resampling that offers sharp results but with the longest processing time. Available as 'separable' and 'non-separable'; the latter gives marginally better results, but is slightly slower than 'separable'.
Original size
Nearest Neighbor
Bilinear
Bicubic
Lanczos 3 Separable
Lanczos 3 Non-separable
Last edited:
Thank you! I had already decided Lanczos was the best for downsampling, but I never saw how much better non-separable is for upsampling.
Here's an example of upsampling with ...
Lanczos 3 Separable
Lanczos 3 Non-separable
It could depend on the image content and the upsampling ratio. Although results are sharper, I've seen the non-separable option produce some artifacts in certain details.
Last edited:
Similar threads
Keyboard shortcuts
- f
- Forum