[JDHill]

Hi again, SJ...the emitter in the provided .mxs is actually set to 500,000 lm. What did you set it to be in your plugin?

~JD

[Maxer]

Great test, although I can't say I completely follow it but I'm glad some of you can and were finally at the point where the discrepancies between V1 and Beta are being discussed openly. I'm also just a user who is eager to get to the end of this road, if this test gets us any closer it's time well spent.

[Nicolas Rivera]

Sorry guys and girls, but how the fuck its this all going to help us all in our working lives???

Will all of this make Maxwell faster and cleaner????

I mean i do not give a shit if maxwell can mimic a rainbow in the eye of a hurracane.... i care about render speed and quality and ease of use.... all of this was promised to us since alpha, and now we have a version thats in some kind of limbo where beta is heaven and RC where hell and V1.0 is the purgatory.

For the love of peace, Next Limit, come up with a better update, a more easy and faster one.

[Mihai]

The Beta shows a much more balanced light-distribution than V1.

Beta shows a much better contrast and tonal detail in the mid-range while V1 looks somewhat flat.

Thanks for making this second test SJ, but I find the conclusions a bit too exaggerated don't you think? To me it's the beta that shows an overall sharper difference between light and dark parts...

Also it is not only gamma that changed with V1, but also the burn value. Should we really make such huge judgements when these two parameters alone can give you entirely different images?

[samsam]

Surely the only way forward for this is for NL to commission someone to re-calibrate Maxwell. I would suggest someone with an artistic eye and who can argue for visual artistic qualities rather than linear mathematical correctness. Ideally a modern day Leonardo DaVini type of person with grounding both in mathematics science and art. Obviously this person has to have the backing of the core programming team as they will need to adapt the code accordingly.

Absolute mathematical perfection is not necessarily the way forward here.

If you think of music for instance the western 12 tone scale is a compromise - it is not mathematically perfect - but we prefer it this way.

Analogue audio - most people prefer audio recordings made using analogue equipment ( we dont like digital distortion at all)

Technicolor - many people prefer the look of technicolor film even though it's highly inaccurate.

Etc;

Many users feel that the beta version of Maxwell had this certain quality that made us stop and look.

I for one am delighted the Ateam is starting to come around to seeing this.

[mmhnemo]

My completely subjective findings with V1:

Light distribution is unnatural.
Beta was better here.

Shadows are too dark and contrasted.
Maybe this is a bounces issue, or engine, or a combination...?

A kind of washed out, blue tinted quality to the renderings.
A few people said beta was too much in the yellow range but not for me - it was pleasing my expectations.

Overall CGish look

[lllab]

i think a very interesting fact is the perceptual truth, like differetn fotopapers have different iamges, or different modern digicams have non linear curves, and each modell shoots a different image.

all of them are unbiased- but still different, some more pleasing than others.
maxwell really should be calibrated like this, to behave not only like a correct camera but alos a nice one.

i am quite confident that this is possible, the problem is more to judge what really make the beat images a bit nicer. it cant be the fact that beta wasnt calibrated correct, it must be another thing- that is my guess.

cheers
stefan

[Mihai]

But you don't take into account at all that materials work differently. In this test, what if the V1 lambert was made just a little darker, and gamma as well as burn was adjusted slightly?

I don't understand the argument of "don't make it accurate, just make it pretty". Some things may seem pretty to you but not to others. You like yellowish tone, others don't.

Everybody should take a look back in the gallery and see all the alpha/beta images that were posted which didn't look too good. Of all the images posted, still only 2% were really good.

[lllab]

No Mihai,

i say make it accurate AND pretty.

each digicam is 99,9% accurate, still each image looks very different.
my thougts very maybe its in these 0,01% that makes perceptual differences.

please keep it unbiased as much as a "normal" camera is unbiased.

cheers
stefan

by the way i am not unhappy with v1 results, still beta looks very nice fro some reason.it is very very subtle differences i mean

[SJ]

Hi again, SJ...the emitter in the provided .mxs is actually set to 500,000 lm. What did you set it to be in your plugin?~JD

I have just checked it: it had exactly the same intensity there.

Thanks for making this second test SJ, but I find the conclusions a bit too exaggerated don't you think? To me it's the beta that shows an overall sharper difference between light and dark parts...

No, look at the pics again and read my observations. The tonal curves are very different - lights and shadows show more subtle detail in Beta (S-shaped tonal curve) while V1 seems to reproduce the "caught" light quite linear. That has nothing to do with global contrast settings. A tonal curve with shoulder and toe produces images with more contrast in the mid-tones because that part of the curve is more steep. That's exactly what Beta does as you can see from the cropped parts. The higher contrast is only there, not at the whole spectrum!

Also it is not only gamma that changed with V1, but also the burn value. Should we really make such huge judgements when these two parameters alone can give you entirely different images?

If I had Burn set to 1 light and shadow areas would have much more contrast as it has now with Burn 0.8. Means: Tonal subtleties would have been ruined more than now.

maxwell really should be calibrated like this, to behave not only like a correct camera but alos a nice one.

"Correct" cameras exist only as long, as no real photo is taken. As soon as you take a photo on film or an image sensor, the light information gets distorted through the characteristics of the sensitive medium. These distortions are either pleasing to the eye or not. I don't know how Maxwell reproduces the light that is caught by it's virtual sensor. Maybe NL could try to embed virtual film profiles - if that is the real cause of the perceived differences.

I don't understand the argument of "don't make it accurate, just make it pretty". Some things may seem pretty to you but not to others.

As stated above, accuracy doesn't exist in photography. It exists physically as long as the light "travels" trough air, is reflected from objects, bounces here and there and makes it's way through the lens. BUT as soon as you catch the light on a medium accuracy becomes obsolete.
And that is exactly what my whole testing is about:
Does M~R reproduce the caught light information different now, or has there been changes in the physical model that affect the visual appearance?
Mihai (and all others who complain), pleeease, I don't want to lead time-wasting arguments based on misunderstanding and subjectivity. I just wanted to find yout what's the core of the problem that people (me included) are not satisfied with V1's render quality. If you find, I made a mistake in my test-settings point it out soberly. If you think it's pointless, skip the thread. If you have ideas and constructive comments building on my tests, writing them down is very much appreciated.

Stefan

PS: testing goes on but sometimes I have to do paid work

[aitraaz]

I would suggest someone with an artistic eye and who can argue for visual artistic qualities rather than linear mathematical correctness. Ideally a modern day Leonardo DaVini type of person with grounding both in mathematics science and art. Obviously this person has to have the backing of the core programming team as they will need to adapt the code accordingly.

Ok then, so you're saying we need to make the call to Buffos...

@JDHILL - Thanks for the help mate. After a long break looking over it again i would say there's enough error on the user end (scale issues, model placement issues etc) to say it would be better to start from scratch (who knows, with the building placed at x 999999999999999 y 99999999999999 z 999999999999 it might have actually been on neptune , so might be a good idea to bring it back to the origin & try again) . Some of the first shots came out pretty well, so i'll give it another go. Anyways justed posted that to try and lighten up the mood a bit

[DELETED]

DELETED

[lllab]

"These distortions are either pleasing to the eye or not. I don't know how Maxwell reproduces the light that is caught by it's virtual sensor. Maybe NL could try to embed virtual film profiles - if that is the real cause of the perceived differences."

yes- this is what i meant.

cheers
stefan

[Mihai]

SJ, when I render your scene it's very dark and I have to raise the iso quite a lot. In the better render, did you have to do this as well, or was the emitter strength already good? Your fstop is set to 200??

[Frances]

When I become dissatisfied to a certain point, I give up.

...as do I. And so we come to the end of our infrequent correspondence.

I guess it's no surprise that you're off my Christmas card list. And I'm sorry things have come to this. The Maxwell Forum used to be my favorite place.

Back to reality - it doesn't sound like this test is an "apples to apples" comparison afterall. The point is not in trying to make the two renders look similar. It is more beneficial to see how the two engines react to the same settings. Even if the settings have different effects between engines. Once a baseline is established, then go from there.

Identical in this case would mean: same camera settings; same lambertian mat with same rgb value; same emitter material; same gamma and burn.