[Class A]

fortran: heh, a physicist, right?

You're right. He could have just been a computer scientist (FORTRAN historically has very good compilers for parallel computation / vector architectures). Falk is probably a bit of both.

[nanok]

you know how they say: it talks like a duck, it thinks like a duck, it walks like a duck, it probably is a duck . i could bet he's a physicist, not saying he's not a computer expert too (most physicists involved in research are, to some extent, anyway)

[falconeye]

CHDK, porting your code to their platform
[...]
on the other hand, your code would be interesting for other people who would like just some nice little solid piece of code to do just that: quick alignment of handheld bracketed shots
[...]
fortran: heh, a physicist, right?

We (ie., my company) haven't decided yet if we will produce a movie preproduction filter. If so, a stabilizing filter would be required, ie., the fast alignment code, after some quality improvements (in the rotation part) which l left out. It would only be after a comparison to deshaker and iMovie that we can decide to release my intellectual property into the open source (otherwise, the code would have shown up in my blog already -- and the other shareholders wouldn't be amused exactly ). But the CHDK idea is in my records now

btw, does the CHDK platform feature enough memory to process three full blown images in memory? K-7 does.

You're right. He could have just been a computer scientist (FORTRAN historically has very good compilers for parallel computation / vector architectures).

i could bet he's a physicist

Didn't I disguise well enough as a photographer?

Actually, I said on several occasions that I hold a PhD in Theoretical Physics. I once was one of Europe's better known computer architects, too. I now run a software and consultancy company. And Pentaxian from first hour 'nough said 'bout me, I guess ...

As for Fortran ... James Gosling (father of Java) some time ago wrote a Fortran parser because Fortran can still produce the fastest code. You wouldn't believe how lightning fast an XML parser can be -- if written in Fortran IMHO, Fortran should be part of the standard repository of every decent computer scientist. I know some who can write Fortran benchmark code but are no physicists or engineers at all. And it will become hard to effectively program a thousand core processor w/o a data-parallel language like Fortran anyway.

[nanok]

We (ie., my company) haven't decided yet if we will produce a movie preproduction filter. If so, a stabilizing filter would be required, ie., the fast alignment code, after some quality improvements (in the rotation part) which l left out. It would only be after a comparison to deshaker and iMovie that we can decide to release my intellectual property into the open source (otherwise, the code would have shown up in my blog already -- and the other shareholders wouldn't be amused exactly ). But the CHDK idea is in my records now

btw, does the CHDK platform feature enough memory to process three full blown images in memory? K-7 does.

i thought there might be more to it than just the impulse of doing it, that's why i tried to be diplomatic in stating the opensource-related idea . it would be cool, but i understand, not everything can be oss, i guess.

chdk as far as i understand is quite varied as far as what hardware is underneath, depending on the camera on which it is running. as there is a whole range of powershots supported, i couldn't tell if all have a buffer big enough for 3 images, or bigger (but i can guess some will have, as there are compacts these days capable of shooting 3fps iirc). i guess that would be essential for speed, but maybe it could be tricked by processing only two pictures at a time somehow, i understand swapping(paging) in and out all the time would kill all real time goodness

Didn't I disguise well enough as a photographer?

aheeem.. well, no. too rational, too straight to the essence of things, typical behavior of somebody used to understanding "exactly how the world works", and knowing when he doesn't understand exactly, that's not something you see everyday, i could spot you from miles .

Actually, I said on several occasions that I hold a PhD in Theoretical Physics. I once was one of Europe's better known computer architects, too. I now run a software and consultancy company. And Pentaxian from first hour 'nough said 'bout me, I guess ...

it might be that i did catch one of those allusions (not outright statements), i am not entirely sure, i might be cheating a bit here . nevertheless, what i said above still holds true: it's written all over you

As for Fortran ... James Gosling (father of Java) some time ago wrote a Fortran parser because Fortran can still produce the fastest code. You wouldn't believe how lightning fast an XML parser can be -- if written in Fortran IMHO, Fortran should be part of the standard repository of every decent computer scientist. I know some who can write Fortran benchmark code but are no physicists or engineers at all. And it will become hard to effectively program a thousand core processor w/o a data-parallel language like Fortran anyway.

tell that to physics students these days "why do they teach us fortran. what the hell is fortran? that's from the sixties. there's c++ now, and what's wrong with visual basic? ****ing dinosaurs..". you don't know if you should laugh or cry, really . being aware of what fortran is, and what advantages it still has is a (cumulative) dead giveaway that somebody has at least something to do with physics.

so to conclude: stay away from direct unarguable statements, don't be afraid to contradict what you just said two paragraphs before, don't make any sense most of the time, bring irrelevant arguments into a discussion and stick to them as if they were the source of all truth, and never ever mention fortran making it obvious you actually know what it is (you can mention it ironically, like referring to some old piece of broken pottery everybody knows is long obsoleted and useless because now there is tap water available, but once it's obvious you actually _know_ what you're talking about, you're tagged), and you'll be fine, probably able to pass for anything, not only a photographer

[falconeye]

so to conclude: stay away from direct unarguable statements, don't be afraid to contradict what you just said two paragraphs before, don't make any sense most of the time, bring irrelevant arguments into a discussion and stick to them as if they were the source of all truth, and never ever mention fortran making it obvious you actually know what it is (you can mention it ironically, like referring to some old piece of broken pottery everybody knows is long obsoleted and useless because now there is tap water available, but once it's obvious you actually _know_ what you're talking about, you're tagged), and you'll be fine, probably able to pass for anything, not only a photographer

Thanks a lot.
I'll stick to this recipe to the minute next time I need to disguise

[dnaseigel]

=Finally, yes, we could probably do anything with a camera, but at a certain point you run into processing issues. We could add full 1080i, 8-channel output, 10fps stills, full HDR with tone mapping, GPS, send e-mails, surf the web, etc, etc, etc, but it would turn into a camera so big and so expensive no one would purchase it.

It is great to know that you could put all of this and more into a camera, but I have to wonder why you think no one would purchase a more expensive camera. With all the other differentiation Pentax offers in their cameras, would it really be criminal to price their products more accordingly to Nikon and Canon?

D3000 w/18-55VR& 50-200VR = $849.99. Has live view and cool AF tracking sequence.

K2000 w/18-55 & 50-200 = $599.95 and every lens gets "VR", uses ubiquitous power supply, etc. (you know the drill)

While this thread may be extremely frustrating, imagine a retailer who is presenting the benefits of buying a Pentax DSLR by showing it's advantages over another brand; when he gets to the part about price the customers almost invariably ask why the camera is so much less? The reaction, at many instances, is such that they make me feel like I am selling them an imitation Rolex.

If Pentax is suggesting that Nikon's cameras are so expensive because Ashton Kutcher is hawking their cameras, Pentax should get itself a pretty boy from Cedar Rapids, IA to be their front man. Unfortunately you would probably end up with some slug like me.

David Seigel

[Class A]

With all the other differentiation Pentax offers in their cameras, would it really be criminal to price their products more accordingly to Nikon and Canon?

David, please don't encourage them. They have started to use the "Gucci principle" on the lenses already.

Your answer why the cameras are so cost-effective (rather than cheap) is simple: Great value for money, no overcharging for a perceived "pro"-image.

I don't see a "cheapo"-image as the major problem for Pentax. Rather their non-presence in advertising media and their non-presence on shop shelves. While the latter may not be universally true, it definitely applies to many areas.

---

Last edited by Class A; 08-05-2009 at 09:13 PM.

[falconeye]

I have done it all the way down. Implemented an alignment operator, benchmarked it, compared its quality for HDR creation to other programs which are commercially available for the task (Photoshop, PhotoMatix, PhotoAcute).
Short result:
The quality of my operator is on par with PhotoMatix and computing the alignment parameters for a 3 sequence HDR takes 350 ms only (on a Mac Mini Early 2009 with one processor used).

UPDATE:

I have updated the algorithm to check out some ideas I had in the meantime.

The algorithm now features subpixel alignment and precise rotation correction (e.g., an image rotated in PS by 10° and ~150 pixels resp. was aligned with +/- 0.3 px alignment accuracy and +/- 0.01° angle accuracy).

Due to a couple of algorithmic tricks the entire operation still is as fast as before (800ms for the 5 image bracket alignment detection on a Mac Mini) and more robust.

The alignment quality now is on par with the best in my comparison -- but much faster ... Interesting. I somehow will have to share it, I guess


So, in camera alignment could be feasible with good quality.

However, as tr13 pointed out, we don't know how well the K-7 processor would perform for plain C code (which isn't supported by a hardware implementation).


EDIT:
I just tried if the algorithm breaks if one image is heavily blurred (as according to a 2007 paper from Microsoft Asia, this is a basically unsolved problem). Well, adding 30 pixel motion blur (in some arbitrary direction) and 10px Gaussion blur (the image looks horrible then ) only introduces an inaccuracy of 0.5px in shift and 0.03° is rotation detection. I.e., it still works great So, it should work fine when computing the panning parameters in a video clip...

[glanglois]

a field in which the noise rather exceeds the signal, alas.

But I'll also admit to learning Fortran and assembler as part of what I thought a decent undergrad education should include. Along with physics, calc, logic, and General Semantics (a type of epistomology, perhaps) but the latter is a different story.

And that was in the '70s.

I'm glad to hear that realists (even from a theoretical background) are willing to speak up. It seems that there's more to computational science than java and VB.

Thanks, doctor!

[nanok]

UPDATE:

I have updated the algorithm to check out some ideas I had in the meantime.

The algorithm now features subpixel alignment and precise rotation correction (e.g., an image rotated in PS by 10° and ~150 pixels resp. was aligned with +/- 0.3 px alignment accuracy and +/- 0.01° angle accuracy).

Due to a couple of algorithmic tricks the entire operation still is as fast as before (800ms for the 5 image bracket alignment detection on a Mac Mini) and more robust.

The alignment quality now is on par with the best in my comparison -- but much faster ... Interesting. I somehow will have to share it, I guess


(...)

this sounds more and more interesting. while you work on making it open source (wishfull thinking, i know ), maybe one interesting idea would be to put it up online somehow, such that people here can upload sample shots to be alligned, and see the results for themselves, on their own shots, and compare to other solutions they are using. just a thought..

[Class A]

The alignment quality now is on par with the best in my comparison -- but much faster ...

Ever thought about offering it to Pentax?

It seems that there's more to computational science than java and VB.

There most definitely is, but the original FORTRAN is even worse. The original FORTRAN is an outdated language design and the only reason why it can be still useful are later language extensions and the availability of highly optimised compilers.

[Samsungian]

Good to know, Thanks

Pentax added a feature THAT NO OTHER DSLR CAMERA HAS! We never claimed it was the "be all" feature for HDR.

A bunch of these threads say "why didn't you do this, and why didn't you do that", when these items would need to be done on a computer post capture anyway. There is plenty of software out there that does this. Why would Pentax, a camera hardware manufacturer, want to start making software?

Plus, we didn't take anything away to add this feature (but we did add enough processing power to handle it which is why it can't be added as a firmware update to the K20D and K10D, so don't even ask). You can still do HDR photography the "traditional" way.

What is the worst this feature will do? Make someone say, "that is really cool, how can I do more? Maybe I'll bracket and use that cool software." I don't see anything wrong with that.

Believe it or not (because I am one of them), there are people out there that don't spend hours tweaking and modifiying images on a computer simply because they don't like to. This gives them another option for image capture.

Finally, yes, we could probably do anything with a camera, but at a certain point you run into processing issues. We could add full 1080i, 8-channel output, 10fps stills, full HDR with tone mapping, GPS, send e-mails, surf the web, etc, etc, etc, but it would turn into a camera so big and so expensive no one would purchase it.

[Nass]

It's a start - it got me started with HDR. Works great on interiors. So it's not too shabby imo.

It's not the whole hog but hopefully it opens up people's eyes to experiment a bit - once they do that they can get into tonemapping etc

[mysticcowboy]

Read the review and agree. I don't want a JPEG file with the camera's guess on tone mapping. If the K-7 were to write the HDR file as an .hdr or .exr file that could be tone mapped in PP, then I'd be thrilled. Or write .hdr + .jpg files for those who are happy with the camera's processing.

On the other hand, Pentax makes the least expensive camera in the K20D and K-7 that allows for a full 9 EV of exposure bracketing. The next least expensive camera that will do that is the Nikon D300. So for a HDR camera the K-7 is still a superb choice.

michael mckee
My Port Townsend – A City in Photographs – 365

[falconeye]

On the other hand, Pentax makes the least expensive camera in the K20D and K-7 that allows for a full 9 EV of exposure bracketing.

It is 8EV (-4 to +4).

[to help reason about this, -1 to +1 would span a 2 EV range, not 3 EV.]

However, I would appreciate the K-7 would offer 3EV steps just like the K-x does (although limited to 3 rather than 5 steps only).
This would extend the bracketing range to 12EV, sometimes required at night.