High Dynamic Range Rendering

Support Compositing and Post Processing
1

Hello all.

I found a really interesting website but i am afraid i need some help with understanding not so much the concept but the process of it all. Would there be anyone who could explain the diagram in simple terms and the four ‘points’ of information under it? I would pay someone if they would like. It seems to have great potential if i could only apply it. I thought simply we load an EXR HDRI image into Indigo or Luxrender, render the image and save it as png or whatever and that would give realistic final result. However there seems to be a little more involved!! Tonemapping, blurring, downsampling ETC. Im not even sure if this is possible to do Luxrender or Indigo? I am looking at doing this for an outside courtyard.

Here is the website: http://www.gamedev.net/reference/articles/article2108.asp

Any sort of insite would be tops!

Thanks :slight_smile:

2

You’re looking at a website that explains how programmers accomplish HDR lighting in a video game. If you want to do it in Lux or Indigo, look for tutorials for those specific programs.

3

I was just about to write back and say that i don’t want this for real time but a single render image. But isn’t the concept the same for Indigo/Luxrender? Do you know where i can find some?

Ta.

4

The concept is the same, but you don’t have to know the technical details of how it works (which is what gamedev.net is explaining). Just find a tutorial that explains how to do it in whatever software you have. Lux/Indigo’s websites would be a good place to look for those tutorials. I’m trying to learn this stuff too, but I haven’t gotten any further than installing Luxrender, before I got distracted by Blending Life.

5

Ok thanks zardos. Yeah because i have seen in places people talking about saving the output as EXR, which is an option in Lux then doing all sorts of thigs to it so it afterwards still maintains that high degree of realism when you convert it back to a LDR image, or something like that, i really have no clue. I got to the point of clicking on the OPEN .EXE output tab in Lux. :o The tonemapping options in these unbias renderers must be more then just flicking through which one you ‘think’ makes your render look ‘nicer’.

Yeah i agree totally, i don’t want to know the tech behind it but i have not found a tute yet. I may try the Lux forums.

Thanks.

6

Are u talking about using HDR to realistically light a 3D scene? Or are talking about HDR for post processing?

7

well blender’s compositor actually has a tone mapping node that works quite well, quite amazingly well actually. It’s not gamma corrected in blender 2.49 though, so that means you have to bla bla bla which means bla bla bla fancy words fancy words.

Basically, blender 2.5 has exactly what you’re talking about. If you leave ‘colour management’ on, and hook up your node setup to a tone-map node just before the final composite output, you’ll be set.

8

Thanks for the help guys.

No i was talking about HDR for post processing…i think, but im still trying to get around exactly what it is im trying to achieve. Could someone explain what this noodle setup is trying to achieve? Why does he use a png along with the EXR image format? What exactly is this doing? I have seen photos taken which are in EXR format that look beautliful, so much better then a standard photo, they talk about combining 3 different exposure setting photos into one. Is this possible with renders? Maybe i should just wait for 2.5. Anyone know when officially it will be released?

Attachments

upload.png661×511 63.4 KB

9

I can’t actually tell quite why he’s done what he’s done, but the guy in your node setup was going for a light bloom effect.

You don’t need different images like that, I shouldn’t think–at least, I never have.

Also, yeah, blender can sort of do ‘exposure’ right now, but not very well. The trick behind the HDR photos you see is not that they are HDR photos, because an HDR photo doesn’t mean that effect you see, it simple means they take record more variance in light than can be recorded with a simple black-to-white scale. The photos you see can’t display any light outside the range your monitor can display, so they aren’t ACTUALLY HDR photos.

What a photographer does to create one of these pictures is to blend several exposures of a picture together, yes, so the darkest parts of the picture are replaced with a very high exposure (making them brighter) and the brightest parts of the picture are replaced with a low exposure (making them less washed out) so you stop seeing so much light and shadow, and just start seeing DETAIL all over.

Blender can kind of change your exposure, but very badly right now, as it only makes the exposure brighter, and only within a very small range. Although there is, now that I look at it, a setting right underneath it (in the World tab under Shading) that does what you want, I think, called ‘Range’. Turning up the range makes it darker and turning it down makes it lighter. In theory you should be able to get a bunch of different exposures out of it.

Blender 2.5 though, handles lighting and stuff a lot better. Blender 2.5x has fixed one very serious problem which older versions of Blender have when it comes to rendering anything realistically, which is very difficult to compensate for without a fair bit of studying. If you’re willing to brave the danger, then look at this. There are some threads around the forum where these things have been discussed, but they’re very old and long, and more places to ask questions at than to try and understand.

10

The comp nodes in the attached image could be any type (HDR or not), they are just adding some nice flares and blooms to the image to make it look nice (or badly shot in camera :wink: ).

A High Dynamic Range image is one that has many more samples between brightest and darkest elements of an image. Modern digital camers often have a range measures in stops of light or contrast ratio. Typically this is quite narrow and results in a loss of information (or detail) in the dark or bright bits.

To overcome this clipping problem (named for the way information is clipped or unresolved by the camera) at the extremities of brightness, you can capture a range of images (typically 3) of the same subject, at different exposures. This means you have an image exposed for the bright bits, the dark bits and the middle. You can combine these normal exposures into a HDR by allocating more bits and smoothing the boundary between each exposure.

EDIT: dang gazumped on explanation, what he said.

11

On a side note what is the difference in using HDR lighting as opposed to Lo colour (8bit image map) lighting?

12

Blender can’t really light with HDR images right now, either. :stuck_out_tongue: Well it can, but it’s not properly prepared to deal with it, so the lighting is extremely noisy.

The theory with lighting by HDR images though is using a picture–or better yet, a light probe, to provide the environmental lighting for your scene–the renderer, in theory, uses the light values of the picture to light the scene.

13

btw has anyone tested the light painting branch from gsoc?

14

Its funny that you mention linear workflow/gamma corection now Zombie John becasue that was a question i asked in the texturing section about a week ago named ‘preparing image textures’ which looked at lowering the gamma on all textures to .45 so as to accomadate the gamma correction on the outoput render. It would possibly seem that what we have been talking about in this thread could be an alternative to this linear workflow? Maybe? Im thinking though that this combining of three images is best suited to actual photos which have been taken in Lo range but where the user wants to combine them to create a high def EXR image in a program like Photomatix because in a Luxrender for example it has an EXR output option, so why would you bother with the 3 seperate images? Wouldn’t you just save the EXR output then simply use Photomatix for tonemapping if u wanted.
Here is a little more info by member ‘tomb’ on Lux forums if interested.

The effect you’re describing is a well known trick to get HDR images out of ordinary LDR (low dynamic range) images where the renderer itself doesn’t support HDR output. Fortunately Luxrender does, so you can import the exr HDR output from Lux directly into any HDR program for postpro. I.e. the steps aren’t needed.

PS. it’s a useful trick for DSLR photography as well, and most modern cameras support it directly by having a mode that takes three images in succession (with different exposure settings) - for example -0.5E, 0EV, +0.5EV - then you can load the resulting images it into Luminance HDR (formerly known as qtpfs, http://qtpfsgui.sourceforge.net/) or some other program with support for LDR series->HDR combines.

What the “correct” balance is is subjective, just as in photography. Unless you’re doing some sort of calibration then eyeballing is indeed the name of the game http://www.luxrender.net/forum/images/smilies/icon_e_smile.gif Use the histogram to make sure you keep within your selected gamut range. Tonemapping is the process where you map a HDR (think “raw” format) image into an image suitable for your display. There are many many algorithms for this, and none of them are “correct”, they just achieve different things. See http://en.wikipedia.org/wiki/Tone_mapping. Also try the different tonemappers in the menu - they are quite distinct. Try importing a raw/untonemapped EXR in the Luminance HDR program - it has a lot of tonemappers implemented.

Types of output: EXR output = untonemapped, raw HDR image, tonemapped EXR image is a LDR image but with high resolution colours (floating point values for R, G and B), the tonemapped png is a ordinary (LDR), low colour resolution (0…255 for R, G and B) image.

If you’re new to Lux, I think personally it’s useful to approach rendering with it as you would photographing a scene http://www.luxrender.net/forum/images/smilies/icon_e_smile.gif

15

oh wait duh, I totally missed the first post. sorry XD

Right, the four points of that diagram are as follows.

  1. Take the hi-res image and scale it down so that it’s easier to work with, and chop out all the LDR values (0-1) leaving only the HDR bits behind (the bits you’re going to bloom).

2-3. Blur the picture vertically and horizontally so that the super-bright HDR bits will be soft and glowy.

  1. Scale the blurred glowy picture up again, and use a Screen or Add effect to blend it back over the original picture. You now have a picture with all the bits too bright for the monitor to display blurred out, so they look all glowy-looking.

  2. ???

  3. Profit!

P.S. The effect I’m describing is not actually what you’re talking about. The effect I’m mentioning is not so much a trick to get an HDR image out of a series of LDR ones as it is an artistic effect created by blending only different AREAS of different exposure, like this or this

16

Its funny how my original question wasn’t what i actually wanted to know and that i was picking up only on hints of what was actually a differnet topic altogether from that first link:o but that bloom effect should look very nice too with a nice tonemapped HDR. So i think the two sort of could go hand in hand anyways. I understand it all now, not quite that bloom effect yet Zombie but i’ll look into and thanks all for helping and sharing your insights. Im going to try a blender render using the ‘range’ tab and see what i can combine using Photomatix. Why manually do it yourself when a renderer has the .EXR output format which would only need tonemapping, well i think you have more control over what can be produced. Thats just my theory. Not sure how all this will apply with Blenders internal renderer but its great to know for Luxrender/Indigo, off the record.

17

OK first off- SUPER sorry about the LOOOOONG post, buuuuut… :slight_smile:

[“On a side note what is the difference in using HDR lighting as opposed to Lo colour (8bit image map) lighting?”]
Excellent question which some never ask! Both will work, HDR has more lighting data to recognize. LDR goes from 0-255 or 0-1. HDR can go beyond that. But Lighting takes the color channels to soooo…
This means that- yes- if you want photorealism- you can in fact use a properly toned LDR photo (regardless of the subject), and it will look realistic.

[“Blender can’t really light with HDR images right now, either. :stuck_out_tongue: Well it can, but it’s not properly prepared to deal with it, so the lighting is extremely noisy.”]
Right you are! This depends on your HDR image. Try rendering a bad HDR in Mental Ray and it will perform the same noisy way. There are ways some software (Such as Mental Ray and Yafaray) compensate that blender does not have as of yet (Photon blurring- not the actual name), but you can compensate for it by performing a blur on the HDR (http://projects.ict.usc.edu/graphics/HDRShop/). Check out that program- allows you to do a WHOLE lot with HDR’s- even the free version…

[“Im going to try a blender render using the ‘range’ tab and see what i can combine using Photomatix. Why manually do it yourself when a renderer has the .EXR output format which would only need tonemapping, well i think you have more control over what can be produced.”]
That would be cool to see how that actually turns out, but its not necessary to render images from Blender and import them into Photomatix to get an HDR. Blender saves out to Radiance (HDR) automatically. Lux does the same thing as Blender. Photomatix is for Photos which is the first way of making an HDR. Renderers like Lux, Blender, Lightwave, Mental Ray, etc. makes higher vlaues (thats right vlaues haha) based of the light settings and scene lights…

Hope all that kind of helps- its just a more flexible format for post that can yield (when done properly) photo realism to a 3D scene. So yeah if you want the max range to work with- then 16/ 32 bit (aka HDR), but keep in mind that it won’t do much for a bad photo, or light a scene realistically if you didn’t have the right photo subject for either. Oh- you can save a single LDR as HDR- it will convert- just won’t have any data on the high end…

Sorrrrrry for the long explanation- but please do check out the simple explanation:
http://projects.ict.usc.edu/graphics/HDRShop (This time click the “Introduction” link)

18

Sorry Zombie John but I am still unsure as to the benifits of HDR lighting? Does it prevent aliasing (banding) from being apparent in a scene lit by a lo-colour image?

19

I’m not sure what you specifically mean by ‘hdr lighting’

HDR lighting just means ‘using a full range of brightnesses,’ from pitch black to white as a supernova, as opposed to just the standard 0-255 fare that most old software used. This can be used in movie renderers, video-games, and newer 2d graphics programs. In this case no, images rendered/saved in this format do not necessarily have less banding–because banding is caused by how many bits each colour channel gets stored with–but since most HDR pictures are saved with their values registered as floating point numbers (1.364, 12.005. 0.0213) instead of ‘0-255’, banding is reduced to the point where it is no longer an issue.

Does that explain what you wanted to know, or were you asking more about lighting a scene with an HDR image?

FWIW lighting a scene with an HDR image is very easy. Basically, the brightness of each pixel is turned into a strength and colour of light, and the picture is turned into like a glowing representation of itself that lights the scene with its own beautiful light, like turning on a TV in a dark room. The fact that an HDR image can store values brighter than ‘plain white’, is a drastic help, because this means it can also use, for example, a picture of the sun to light the scene, improving its lighting effectiveness from a TV screen to a window.

20

ZombieJohn is correct in his explanations.

There is nothing that would prevent to store HDR data in an 8-bits image format. But the 8-bits image format would produce very noticeable banding in this case. The human eye will perceive the banding is the difference between two tones are beyond a certain treshold. For this reason, an 8-bits image can store about 7 f-stops of illumination range before the banding become noticeable.

In comparison, OpenEXR format can store something like 40 f-stops of illumination range.

Camera RAW formats can store about 10 to 11 f-stops and using the autobraketting feature of some cameras, one can get about 13 f-stops in 3 exposures.

Using a HDR environment map to light a scene is called Image Based Lighting (IBL). Although any image can be used, including any LDR image, it is only with a HDR image that one can get the required dynamic range that would be found in a normal scene. Even when the aplication does not support IBL, using a HDR image for reflections on a dark surface can give reflection effects that would not be achievable with HDR maps.

Linear Workflow is something else completely and is only related to HDR by the fact that HDRI and LDRI have different color space. The HDRI is already in linear color space while the LDRI is most probably in sRGB (gamma 2.2) color space and ths needs to be taken into consideraton when working with a linear workflow.