Creating an HDR - F-Stops

[jujubee]

I'm trying to create a bunch of .hdrs that I would be more than happy to share with everybody. The only problem is that I don't completely understand my digital camera nor some of the more technical terms in Debevecs creation tutorials when it relates to bracketing.

I figured out how to use my digital camera to automatically take a series of different exposures. The only problem is that I'm not confident with what the numbers mean. For example my auto bracketing settings gives the following options:

+/- 0.3
+/- 0.6
+/- 1.0

These three options can be set to x3 or x5. I take it that these are additive values multiplied by the number of pictures to be taken.

My default f-stop is 2.8. This aperture value does not seem to be changing (according to my lcd) when I'm running a series of exposures, although the shutter speed is jumping from 1/30 to a seeming 1/300.

So given these settings, is it safe to assume that 5 images taken at +/- 1.0 are going to jump from 2.8, to 3.8, ~7.8 respectively?

[vance3d]

I'm going throught the same thing now with my digital camera.

As I understand it, many cameras today use an exposure value system (EV)

EV's are combinations of f-stops and shutters speeds at set film speeds.

its possible to have the value of EV 1.0 be a wide variety of f-stop and shutter speed combinations.

by bracketing in whole increments of EV you are basically increasing or decreasing the exposure value by whole fstops.

whole f-stops are 1 1.4 2 2.8 4 5.6 8 11 16 22 32 etc.

What your camera might do to achieve the bracketing is to hold the f-stop and adjust the shutter speed thereby increasing or decreasing the amount of light hitting the ccd.

so if your camera reads f2.8 at shutter speed 1/30 then increasing the EV by 1.0 would I think change the shutter speed to 1/15 and decreasing the EV by 1.0 would change the shutter speed to 1/60

my camera seems to have a limited range of f-stop settings but a wide range of shutter speeds so I'm still able to take a wide range of exposures by adjusting the shutter speed alone.

I hope this helps you as its been a while since I've had to deal with f-stops and the like.

V Miller

[J_Bug]

Can I ask what cameras you both have? I've been looking into getting a digital camera that does bracketing.

Thanks,
--Jon

[richard_rosenman]

Hey guys, I'm also doing this as of late and here is some info I have found particularly useful at the Debevec forum.

Should should NOT change your f-stop for the different shots, but rather change your exposure values. The reason for this is because f-stop changes your depth of field and this is undesirable from shot to shot.

I have successfully achieved a good camera calibration curve (almost completely linear) in HDRshop by shooting at 1 sec, the 1/2 sec, then 1/4 sec, then 1/8 sec, then 1/16 sec, then 1/32 sec etc....

I also tried with a 10 inch gazing ball I purchased at my local hardware store but found that it is far from perfect as a sphere and caused a lot of distortion.

Hope this helps,
-Richard

[jujubee]

Thanks Vance for the info- perhaps its just grabbing the EV's then. I guess I just have to manually translate these shutter speeds into f-stop values which hdrshop asks for. Judging by what you have posted Richard (which makes absolute sense), it's kind of counter-intuitive that hdrshop asks for f-stop information but that is very good to know.

As for the camera calibration curve, I het have yet to do this. That's next on my list. It would be great if all of us hdr enthusiasts could assemble a listing of calibration curves for specific camera models in this forum. This would take a chunk of work out of the process. From what I've read by Debevec, almost all digital cameras read in as linear. I guess we'll find out.

I also purchased some 'christmas balls' today at Target. I'm sure they're not perfect distortion-wise- but they're great (and cheap) for expirimenting. Funny enough I couldn't find anything at Home Depot- the closest thing I found was a cabinet door knob with a distorted base. Don't waste your time going there as the also don't sell any bearings. Then I had thought perhaps a trailer hitch would do the job- the thing I discovered at a marina is that their top is lopped off...

Here's probably your best bet if you're looking to get serious:

http://www.mcmaster.com/ - do a search for 'steel balls.' There's a 2" diameter listed for $17 US. The 3" jumps to $100. I called around and found other places that produced these bearings- they were roughly $50 for 2" so this is the cheapest I've found thus far. Most companies do not produce beyond 3".

The other aspect to creating panormas is the fact that every camera is going to have some level of lens distortion- and it will add up noticeably after stitching many images together. A couple years ago I experienced this using Stitch- and couldn't get the values exactly precise. My 360 degree panorama ended up having an old 'kung-fu' type of distortion applied to it. I wonder if the Photoshop plugin Lensdoc would take care of this 'bubble' effect.

Here is an interesting guy with alot of good links. http://www.gregdowning.com/HDRI/stitched/

Personally, I think he went way overboard with the amount of shots he processed though the results are pretty impressive. I think he would've been much better off time-wise with a simple christmas ball...

As for the camera I am using J_Bug, it is an Olympus C3030 Zoom. I absolutely love the image quality from Olympus- it even has support for external flashes which is something alot of digital cameras lack. Will post the values when I get around to it later.

[J_Bug]

Great thanks

[jujubee]

Funny Richard- youre on the same wavelength but several years and days ahead. I found your other f-stop posting. I also found this info in a thread at debevec's website- it's a good read but a wee bit heavy on my brain- still not sure what it's trying to tell me as my math sucks:

http://pub86.ezboard.com/fhdrshopfrm...icID=100.topic


And a little bit off-topic, but I'm lost on removing the photographer. It looks like this guy is really bright and according to the pictures his techniques works wonders, but his english is confusing the heck out of me. Can anyone translate this technique?!?

http://www.cgtechniques.com/tutorials/getridof.php

[jujubee]

Ok. Now that you're thoroughly confused by the matematical ninnies in the above ezboard link and by people who confuse f-stop with aperture and shutter speeds, here is the simplified formula as to what 'I' think needs to be done on a digital camera:

a) Keep your aperture always the same. So my default is 2.8. Ill leave it at that.

b) Manually snap 10 pictures (just an arbitrary number I'm making up and one that keeps my f-stop steady at 2.8- depends on what kind of quality you desire for your modeling needs as well as your camera's capabilities- perhaps 10 is overkill) at the shutter speeds from the following range (they will differ slightly according to your readouts so these are approximate values- my camera's readout appear on the right in correlation to actual values on the left):

1/1024 or 1/1000
1/512 or 1/500
1/256 or 1/250
1/128 or 1/125
1/64s or 1/60
1/32s or 1/30
1/16s or 1/15
1/8s or 1/8
1/4s or 1/4
1/2s or 1/2

If you notice- these are all factors of 2. According to what I've read from the following statement written by Debevec and Mallik, this would mean that each one of these shutter speeds would correlate to roughly 1 f-stop value in hdrshop:

"By analyzing recordings we were able to verify the accurracy of the exposure times to within a thousandth of a second. Conveniently, we determined that the actual exposure times varied by powers of two between stops (1/64, 1/32, 1/16, 1/8, 1/4, 1/2, 1, 2, 4, 8, 16, 32), rather than the round numbers displayed on the camera readout (1/60, 1/30, 1/15, 1/8.... etc)"

So guys I hope I'm not completely off-base here with my assumptions. It makes sense to me thus far so now I'm going to give it a run and see how it turns out. Please write me if you change these settings or you think I'm completely wrong.

[jujubee]

Richard- I just calibrated my curve. Initially I got a wild graph and I thought something was completely wrong. I pulled in the markers and resolved until I got to a place where the curves no longer diverged greatly- and got a line very close to 1.0. The instructions really didn't tell me much about having to do this but I am assuming this is the correct method. Did you experience something similar?

[richard_rosenman]

Hi Juju

I am following EXACTLY the same procedure you just outline with regards to the exposure times and f-stops.

I got a really nice curve right from the start. So it wasn't tough to get to the same linear curve Debevec has in his tutorial.

I'm going to try several more tests and let you know my results. By far, the most difficult part for me has been removing myself from the photograph - the procedure is extremely complicated and extensive.

-Richard

[jujubee]

Just a note to some digital camera users (in particular the Olympus C series), I believe more than anything that you will want to set your camera to manual and avoid bracketing altogether. My camera has a switch called 'a/s/m', which stands for aperture, shutter, and manual. Once set to manual within the submenus, I am able to adjust my shots according to the chart I posted above without automatically altering f-stop values. In shutter mode my F values were jumping slightly so that doesn't seem to be the answer.

There are three numbers at the top of my camera lcd. The first (from left to right) is f2.8 (f-stop), the second is my shutter speed at 1/2, and the third is one that automatically changes called exposure compensation (jumps from values of 0, .3, .7, etc...)

I talked to a rep at Olympus today and he mentioned that exposure compensation (not to be confused with exposure value or EV's) is a value which is used for reference by you in this mode- not a value that adjusts your camera. Exposure compensation attempts to let you know when you're camera is set up for best lighting conditions. So in a nutshell, you may want to ignore this value.

I'm going to do some more tests later on a variety of different methods (including bracketing) to see what the end results are. I haven't quite reached the point of having to remove the photographer but I take your word on it Richard- was just reading ahead to get a good idea on the entire process. If you manage to decipher that guys removal techniques- please let me know how to do it. I wonder if this is easier (and more accurate) than Debevec's technique of using masks in Photoshop. Even further down the road- I was wondering if you think there's any Photoshop actionscripting that could be applied to this method instead of having to do this manually every time.

And one last question for now- I was wondering if you obtained and chromatic abberations around the edges of your spheres. And if so, did you use any good methods to get rid of these colors. I saw that one person used a Photoshop plugin to lessen these effects.[/b]

[jujubee]

Guess it's impossible not to obtain some aberrations... Anyhow- Not sure if you're doing this Richard but I think it was much easier for me to assemble the .hdr AFTER previously shift cropping the images in Photoshop while viewing them in the measurement window for consistency. I got a lot less blur then afterwards. Unfortunately I should have shot my images closer as well as not used a tissue to wipe off my sphere.... If your measurements are off from the crop and you you try to assemble then hdrshop will crash when trying to calculate your .hdr. I wasn't too certain which method I should use- by specifying f-stops or trying to automatically calculate. I take it automatic calculation should be able to determine even closer shutter speed ranges for a better overall image. I'll have to test that later...

[jujubee]

Another update. I finished my first light probe and will finally test it in a model! And beyond all the trials and tribulations, it's been quite fun to have a better understanding of this entire process. The probe is rather embarassingly poor so I'm not going to post the hdr. After all, it was my first time creating one. Masking out myself was not an easy task but then again I will try to shoot better/cleaner images next time around. In my tests I found out the following for masking out myself and the probe/camera stand- the farther away you take your pictures from the sphere, the less you have to mask out yourself afterwards. But this comes at a price of higher quality versus efficiency unless you have a great zoom feature or a high enough camera resolution format. My mask on the other hand is causing strange discolorations where the blend occurs. If you know how to get around these artifacts, let me know. Perhaps this is what you were refering to problem-wise.

And on a separate note, doing this with a reflective sphere was a ton easier then having to stitch a series of photos together. Just something to keep in mind next time you decide to shoot a VR Panorama...

[jujubee]

And here's another update. The results of my first .hdr:







Well, it works for one thing. My dining room hdr definitely effects my models. Next time I have to take better pictures. This is a trial and error process.

What it tells me is the following:

1) the height of which you capture your image at is basically the same height at which your models would appear. Adjust accordingly. Perhaps obtain a table tripod for outdoor scenes. Will have to test how much it distorts a ground plane. Perhaps there is a magic height number which camera lenses are adjusted for- thereby decreasing distortion when properly leveled.

2) Don't use tissues to wipe off your Christmas Ball. You can actually see the dust on the first sphere image from the tissue lol... It pays to have a microfiber or a computer dust blower handy.

3) Don't use a christmas ball if you are going to get serious about quality. Buy a chrome polished sphere online. For basic purposes this will work fine.

4) I have to find a balance between shooting closer while making it easy enough to mask out myself. Farther distances make masking less of a chore.

5) Use Tiffs from the camera instead of other formats. Tiffs will not show compression artifacts...

6) Attempt to get a higher dynamic range - shorter increments of shutter times will get auto-loaded/calculated by hdrshop. Just check your values to see if the brightness in hdrshop follows an incremental range. And the more images the better. Going by 1 f-stop increments of 1/2 by powers of two, my camera maxes out at 1/800 which is not the 1/1000 which I desire. This forces me to stop at 1/500. My first images at lower shutter speeds were washed out by white from the light. These had to be deleted in assembling the .hdr. It ended up forcing me to only use 5 photos for each 90 degree angle.

7) Zoom in as close as you can get to your sphere- haven't tested out my camera's digital zoom function yet to see if it does anything funky. Not even sure if it works in manual mode. To be determined.

See if Macro mode is supported on your camera- find the balance between macro and the distance away from your sphere. determine what produces the best quality (on or off.)

9) Save your pictures at the highest resolution possible by your camera.

10) the images above are suitable for lighting conditions. They could be better. I wouldn't go so far as to use these hdr images in my viewports- I'd use them in the vray environment instead and overide the Max render display. Using them in this manner, it is possible to use just about any quality of hdr to light a scene. But more dynamic and larger hdrs convey nuances and a sharpness in color differences that low quality hdrs do not convey.

Hope this helps...

[jujubee]

One more note to anyone interested as I had through some messages here. Debevec states ge had used a 2" chrome sphere in pretty much all of his scenes. A 3" spehere was used for Fiat Lux. I think the bigger the sphere you have, the greater the chances of distortion. On the other hand you may get a better resolution. My Christmas ball is roughly 2 inches in diameter.