Genuine Fractals aka perfect-resize7 for animation

[TheBadger]

Using "Tower Tutorial" by "schmeerlap", available here: http://forums.planetside.co.uk/index.php?topic=12630.0
I created a simple animation to test in "perfect resize7".

Original animation
432x243
Detail: 0.8, A:5, GI: 3, 2, 8
No other changes were made

The animation took 2m30s per frame, for a total 3h17m14s to finish at 100 frames total.
Here is the un-resized original animation: http://vimeo.com/27063162

I then resized every frame in perfect resize to 1280x720 by batch process, batch processing takes very little time.
Here is the resized video: http://vimeo.com/27063337

As you can see, the resized image shows the lack of detail in the original much more clearly, but resizing, in and of its self, does not degrade quality to a high level.
It seams that if a animation render is done with high detail but small pix ratio, then the animation can be resized while maintaining high quality.

There was no additional flicker, though the flicker that was there became clearer.
The smaller the resize the better the results. I think that this is a viable way to save many hours rendering

My vimeo account is not premium, so it is not in HD, but you can still confirm my test.
Thanks, hope you will chime in!

[Henry Blewer]

This is an interesting test. I would like to see how objects are treated using this process, and with higher detail settings. I did not notice any artifacts caused by the resize.

[Kadri]

You made me curious TheBadger!
I am rendering now the same scene with  higher settings too.
I want to see it without flickering if possible to judge the resizing better.

[freelancah]

The result is quite impressive. I had quite a bit of doubts about this but it seems to be working very well, nice!

[TheBadger]

Because my vimeo account is not in HD, the HD video I uploaded was compressed down to whatever the max size a free account can have. So, The resized video is again changed back down in resolution, and it still looks nice:D

Anyway, so that everyone can see detailed results, here are frames from each videos to compare.

To be clear, I think my upsize from 432x243 to 1280x720, it very extreme. I would not make such a large jump on a real project. As I mentioned above my render time per frame was only 2min30s. I could easily live with twice that, I would even be content with 10min per frame, this would make a much higher quality video to begin with. The detail must be there to begin with, or the resize is pointless!

[TheBadger]

I am rendering now the same scene with  higher settings too.

Kadri, please post your results! Everything in terragen is about balance, but I haven't found a balance with settings, my system, and my expectations yet. It would be great if you are able (or anyone) to see more clearly a path to make resizing a real tool for us to have. And post it!

I will also do this test with adobe "darkroom3", and "VideoEnhancer". (free test versions)

Thanks

[neon22]

Please also be aware that when you render animations there is a Camera Blur setting which simulates motion blur. It is likely that this will have significant effect on the upsizing quality.
I suggest you try with and without motion blur factor. I.e. 0.0 and 0.5 to compare.

Specifically the 'Super Resolution' approach is affected by this. http://www.infognition.com/articles/when_super_resolution_doesnt_work.html

Also I would expect to see artifacts in the image using Perfect Resize 7 - the blur artifacts may become strengthened into more solid lines - which may be visually distracting.

So you may also wish to consider a slow camera motion and a fast one in your tests.

Good luck with your efforts. looing fwd to the results.

[TheBadger]

Specifically the 'Super Resolution' approach is affected by this. http://www.infognition.com/articles/when_super_resolution_doesnt_work.html

Hello Neon,

My question then is, is upsizing using an Algorithm the same as compressing using a Codec?!
Are Algorithms and codecs two names for the same thing?

expect to see artifacts in the image using Perfect Resize 7 - the blur artifacts may become strengthened into more solid lines - which may be visually distracting

can this problem be avoided by not compressing, or are you saying that an image is compressed during the process of resizing?

[Kadri]

... is upsizing using an Algorithm the same as compressing using a Codec?!
...

I am not a programmer but Algorithm's could be used anywhere in programs and filters etc. like videos and like TG2 too.
The thing is that upsizing and compressing do use some math but are different .
Upsizing does make the picture bigger (as we try here) obviously and compressing with codecs do make the video smaller (in megabyt , gigabyt etc) for use in any kind of media you want to use them.The video would be too big to use in standart media without a codec(compression) .
Most users do think that videos that have extensions like Avi or Mov etc. are codecs but they are only container formats.
They can have many kind of codecs used in them for the video they contain(!) (xvid , x264 , mpg ... ).
It looks a little confusing at first but when you begin to research it will be easier to understand in a short time.
http://en.wikipedia.org/wiki/Codec
There is a little very nice but unfortunately in the last years not much developed program:
http://en.wikipedia.org/wiki/GSpot
And as you see there you can use this too  to see what codec is used in the videos you have.
http://mediainfo.sourceforge.net/en

[neon22]

My question then is, is upsizing using an Algorithm the same as compressing using a Codec?!
Are Algorithms and codecs two names for the same thing?

A codec uses algorithms to do work.
An Algorithm just describes a way to do something.
A Codec is specific to taking data in one form and transforming it into another. Generally just used for multimedia streams. It comes from the word encoder/decoder and implies it can go in both directions (not always true).
Wikipedia is your friend here...

The main problem that you are trying to solve is to do with information theory and signal to noise ratios.
The problem could be rephrased this way:
"What is the lowest resolution you can render at and scale up, in order for the end result to look visually indistinguishable from a 'true' result."
True - in this sense - referring to a full resolution render.

Clearly you will lose information rendering at a lower resolution but the human visual system has some weak points which can be exploited by clever algorithms.
Amongst these aspects:

• We see edges formed from luminance (e.g. in grayscale) quite a bit better than we see those from just colour.
•     hence video is frequently compressed unevenly in the luminance and chrominance domains. Lookup 4:2:2 as an example. Common in TV codecs.
•     you can see the resize tools trying to maintain the visual edges (in your prev upscale examples)
• Dynamic range. Turns out we can use low spatial resoultions if we have higher dynamic range.
•     we used to print film at 2kx2K at 14 bits instead of 4kx4k at 8bits. even 1kx1k at 12bits was acceptable for a while.
•     rendering to openEXR format and then correcting to form a tone mapped image in a paint app is a useful way to control the exposures in a rendered image.
•     One day everyone will do this. Checkout the light-exposure control in Camera nodes
• Motion blur. Simulating the effect on your retina of motion in the scene. When things move quickly we perceive less of them.
•       you can render at lower res if you never slow down 
• Depending on our distance from the image - we need to consider "circle of confusion"
•      this leads us to DPI. Posters designed to be looked at on a wall do not need to be as high in resolution as those viewed in a book. The closer you get, the more you can see the gaps in between the pixels. This principle is frequently not well understood but also does not apply here as the output format has already - probably - been calculated for you.
•      E.g. for PAL video with square pixels, the calculations have been done and its 768x576

The signal to noise tradeoff is more to do with how compressed you can make the encoded files. This only matters if you need them small for transmission or if you have a 500Kb max image size upload to a (say) website 

I don't expect anyone cares to but information theory has lots of academic stuff to say about this subject...

The end result is the question as stated above.
Tradeoffs are soft edges, new false edges, and loss of detail in slow moving sections.
Always process the highest quality original imagery you can (so not jpegs, which already have edge artifacts added and these false edges will be enhanced)

[Kadri]

...
I don't expect anyone cares to but information theory has lots of academic stuff to say about this subject...
...

You sound like you know 1 or 2 things , Neon22   What is your profession if i may ask?
There are too many links ... Can you give some links for the " information theory " subject that you think are good neon22, preferably without much math?

[TheBadger]

This only matters if you need them small for transmission or if you have a 500Kb max image size upload to a (say) website 

lol, yep. I was thinking of asking staff if they would let us post the high res tests, so that everyone could really see whats going on in the results...after all the tests are in, and assuming they reveal a benefit to the terragen2 user. (Whenever someone complains about render time link them here.) But I don't know how interesting this is to anyone else, other than those of us who are participating in this thread. It might just seem like tricks and gimmicks to some. Maybe in the end it will be? But good results so far.

Also,

"What is the lowest resolution you can render at and scale up, in order for the end result to look visually indistinguishable from a 'true' result."

I like that in perfectresize7 you are able to work on one image, make all your changes and save a preset. Makes answering the question a bit easier.

Going to try lightroom processing now, will post when its done.

[TheBadger]

Lightroom3 test:

Lightroom3 does appear to do a better job upsizing than perfectresize7

See here and go to bottom of page: http://www.have-camera-will-travel.com/field_reports/perfect-resize-7-pro-vs-ado/

However, lightroom3 does not except .bmp file type, or .exr and using .jpeg (as stated above) would be counter productive.

Lightroom 3 supports more than 290 native camera raw file formats, in addition to DNG, TIFF, and JPEG files — in other words, the formats primarily used in digital cameras. In addition, Lightroom 3 also supports the Photoshop PSD file format for enhanced integration with Photoshop CS5. For a complete list of manufacturers and models supported in Camera Raw, see the camera raw page.

If one wanted to, it is possible to convert each frames file extension to .psd or to (output animation from terragen2 with .tiff (correct?).
But even if you do this, you are only able to resize when exporting from Lightroom at the end of the process, this would be fine but for the fact that sharpening the resized image would have to be done blind, or in another program making the workflow even more cumbersome. For a detailed explanation of the process see here: http://www.digital-photography-school.com/how-to-resize-images-in-lightroom-2

Test result: Fail- Not practical.
If we are only replacing render time with a convoluted workflow, I don't see how the situation has really improved.
Additionally, perfectresize7 has built in presets for all standard video sizes including HD, one need only render out of terragen in the same aspect ratio as the target size and then  visually apply whatever sharpening and effects you like, save a new preset, batch process, and your done.

Next test, "VideoEnhancer": http://www.infognition.com/VideoEnhancer/

[TheBadger]

Video Enhancer:
Windows only, sorry cant test. looks nice though, particularly if you are an editor/compositor.

Will continue testing perfectresize7 this week with better methods.

Thanks

[Kadri]

Lightroom3 test:
...
If one wanted to, it is possible to convert each frames file extension to .psd or to (output animation from terragen2 with .tiff (correct?).
But even if you do this, you are only able to resize when exporting from Lightroom at the end of the process, this would be fine but for the fact that sharpening the resized image would have to be done blind, or in another program making the workflow even more cumbersome.
...
Test result: Fail- Not practical.

I have heard much more cumbersome methods for things in the 3D world
I think it is not so bad especially if you count the render time for good looking renders in TG2.