ProfHankD

ProfHankD

Lives in United States Lexington, United States
Works as a Professor
Has a website at http://aggregate.org/hankd/
Joined on Mar 27, 2008
About me:

Plan: to change the way people think about and use cameras by taking advantage of cameras as computing systems; engineering camera systems to provide new abilities and improved quality.

Comments

Total: 696, showing: 141 – 160
« First‹ Previous678910Next ›Last »
On CP+ 2015: Sony shows off new technology article (216 comments in total)
In reply to:

crashpc: No new breathtaking sensor? Whuh, it seems that we´re stuck at 24Mpx for APS-C and 50Mpx for FF. :-(

crashpc: Look at lens resolution measurements (made without a sensor) -- they rarely get that high even in the center at the best aperture... and the published resolution figures are quite consistent with what I see on my meager sampling of 140+ lenses. ;-)

As for the 40MP Olympus, it's doing multi-image superresolution processing -- which is not subject to the same limit. In general, you'd be surprised at how much more detail can be computationally recovered (especially detail that's really just credible synthesized guesses). As I explained, most lenses are already being sampled past Nyquist under most circumstances, so the improvement is tiny -- but not zero -- as you increase sensor resolution further.

Don't misunderstand me: I'm all in favor of increased resolution, I'm just saying that little physics issues like the fixed wavelength of light and diffraction effects make really high resolution a thing for bigger sensors. In fact, I'm working on a 500MP 4x5-format sensor.

Direct link | Posted on Feb 15, 2015 at 14:37 UTC
On CP+ 2015: Sony shows off new technology article (216 comments in total)
In reply to:

crashpc: No new breathtaking sensor? Whuh, it seems that we´re stuck at 24Mpx for APS-C and 50Mpx for FF. :-(

28MP from Samsung for APS-C.

No matter. Very few lenses come close to challenging 24MP APS-C or 54MP FF (yeah, a few more can challenge Canon's 50MP FF in the center). Increases in sensor resolution beyond the Nyquist sampling of the lens resolution still helps, but way less than before Nyquist, so I doubt that even a jump from 50MP to 500MP in the same size sensor would show much improvement across the frame with the vast majority of existing lenses. Realistically, I'd expect the major benefit to be a touch less degradation when correcting for lens distortion.

BTW, lenses for larger formats still have about the same resolution per mm, so a larger sensor can productively go much higher. See http://aggregate.org/DIT/LAFD/

Direct link | Posted on Feb 15, 2015 at 13:05 UTC
In reply to:

ProfHankD: "Sensor manufacturers have concentrated mostly on providing high ISO settings that are not often used, he said, and had neglected low settings in their favor, but Olympus hopes this will change very soon."

As QEs go up, it gets harder to have a big enough place to dump all that charge. The problem as I see it is sensors trying to use the same integration time for all pixels. For example, why not have a bright pixel sampled multiple times, and then averaged (digitally), during an exposure long enough to get detail in the dark pixels? I've been doing a better variant of this in my research: http://aggregate.org/DIT/ei20140205.pdf

Hmm. Slow due to a badly-coded median filter? 30-pixel window processing shouldn't be that slow... but my background is in parallel supercomputing, so I might be a bit more careful about coding efficiency than DxO is?

Direct link | Posted on Feb 15, 2015 at 03:54 UTC
In reply to:

ProfHankD: "Sensor manufacturers have concentrated mostly on providing high ISO settings that are not often used, he said, and had neglected low settings in their favor, but Olympus hopes this will change very soon."

As QEs go up, it gets harder to have a big enough place to dump all that charge. The problem as I see it is sensors trying to use the same integration time for all pixels. For example, why not have a bright pixel sampled multiple times, and then averaged (digitally), during an exposure long enough to get detail in the dark pixels? I've been doing a better variant of this in my research: http://aggregate.org/DIT/ei20140205.pdf

More than likely what you're really thinking about is DxO constructing a noise model for the image... and yes, that's not computationally cheap. Fortunately, we don't need to do that just to avoid saturating bright pixels.

Direct link | Posted on Feb 15, 2015 at 02:24 UTC
In reply to:

ProfHankD: "Sensor manufacturers have concentrated mostly on providing high ISO settings that are not often used, he said, and had neglected low settings in their favor, but Olympus hopes this will change very soon."

As QEs go up, it gets harder to have a big enough place to dump all that charge. The problem as I see it is sensors trying to use the same integration time for all pixels. For example, why not have a bright pixel sampled multiple times, and then averaged (digitally), during an exposure long enough to get detail in the dark pixels? I've been doing a better variant of this in my research: http://aggregate.org/DIT/ei20140205.pdf

Actually, it's mostly a matter of fancier control logic for the sensor; the extra computation is easy. To do this with minimal mods to a conventional sensor, the "averaging" I was talking about literally just becomes summation that could easily be done as fast as the sensor readout. Don't know what averaging DxO-9 is doing, but such operations on the raw buffer inside a Canon PowerShot (compiled ARM32 code put inside the camera via CHDK) happen faster than writing a raw file.

Direct link | Posted on Feb 14, 2015 at 21:35 UTC
In reply to:

ProfHankD: I'll repeat my standard plea: wider, smaller, slower, cheaper, ultrawide zooms. I don't care if wide open is f/5.6 (or any number on the good side of the diffraction limit)... but I'd like it to start at 12mm (for FF) and be smaller and much cheaper without having poorer IQ. I have faith that somebody should be able to make a nice 12-24mm f/8 to sell for $300... it can even be manual focus... anybody?

"Real" == Antique? I can't imagine why anyone wants an optical viewfinder... but then, I have had the unfortunate experience of actually using OVFs for four decades, so I'm clearly biased. ;-)

Direct link | Posted on Feb 14, 2015 at 20:39 UTC

I'll repeat my standard plea: wider, smaller, slower, cheaper, ultrawide zooms. I don't care if wide open is f/5.6 (or any number on the good side of the diffraction limit)... but I'd like it to start at 12mm (for FF) and be smaller and much cheaper without having poorer IQ. I have faith that somebody should be able to make a nice 12-24mm f/8 to sell for $300... it can even be manual focus... anybody?

Direct link | Posted on Feb 14, 2015 at 17:31 UTC as 4th comment | 5 replies

"Sensor manufacturers have concentrated mostly on providing high ISO settings that are not often used, he said, and had neglected low settings in their favor, but Olympus hopes this will change very soon."

As QEs go up, it gets harder to have a big enough place to dump all that charge. The problem as I see it is sensors trying to use the same integration time for all pixels. For example, why not have a bright pixel sampled multiple times, and then averaged (digitally), during an exposure long enough to get detail in the dark pixels? I've been doing a better variant of this in my research: http://aggregate.org/DIT/ei20140205.pdf

Direct link | Posted on Feb 14, 2015 at 17:17 UTC as 49th comment | 8 replies

The air is the only interesting-to-me thing here, and it's specless. Is it more open than the Sony units that probably will give superior IQ in the same form? What I want is CHDK-like programmability INSIDE the camera... does air even suggest it will do that? The wireless tethering is cute, but Sony's there first with a now fairly decent and very portable API.

PS: I love how the diagram shows a supposedly 3D-printed prop stuck on top as though that would work with no motor, no power, and no controller (and apparently no way of using the air as the controller). They're right that the market isn't folks who want to stick these things on their cell phones, but they sure are clueless about what camera hackers (an 3D-printing makers, of which I'm both) really want....

Direct link | Posted on Feb 14, 2015 at 16:56 UTC as 8th comment
On CP+ 2015: Canon shows off prototype 120MP CMOS sensor article (255 comments in total)

"Note the distinctly different architecture" -- I'm sorry, but my ability to see sensor architecture in a photo like that is no better than my ability to read off the order of cards in a deck by staring at the deck. ;-)

All I see here is Canon's continued claim that they are still a "leader" in the sensor manufacturing business... ok, if they say so. 120MP in a package smaller than full frame is significantly over Nyquist sampling for every camera lens I've ever seen, so what's the real message? In fact, when Intel was showing lots of chips like that, what it really meant was that their yield was terrible, so they had lots of duds to creatively discard... is that what Canon is doing too? Anyone see any evidence that Canon has produced even one fully functional 120MP sensor? Sadly, I honestly don't care....

Direct link | Posted on Feb 13, 2015 at 21:28 UTC as 18th comment

Well, it's nice to see I'm not the only one 3D printing camera prototypes... but this doesn't give me warm fuzzies about when we'll see the actual product. This reeks of being only a marketing test. Let's hope I'm wrong. ;-)

Direct link | Posted on Feb 13, 2015 at 02:11 UTC as 53rd comment | 2 replies
In reply to:

ProfHankD: I don't think Canon is very afraid. I think they're mostly trapped in a development cycle that keeps most subsystems unchanged across many years, upgrading only one or two at a time. This year, apparently susbsystem they are most proud of is the new mirror drive mechanism... which isn't helping them make a more viable mirrorless product!

Anyway, the EOS M3 is headed in the right direction, but incremental evolution hasn't been fast enough to keep up with Sony or Samsung. No point in releasing the M3 in the US where it would probably hurt sales by effectively blessing the approach better executed by Sony, etc. The lack of native lenses isn't a big problem in reality, because the M models all support adapting EF and EF-S lenses (Canon couldn't re-engineer more than a few per year as native EOS-M mount anyway).

BTW, I have an EOS-M, and overall it is severely outperformed by my Sonys (even the original NEX-5), but I have it because it can be reprogrammed using Magic Lantern (ML).

rrccad and Vignes, the 7D II doesn't have a full-frame mirror -- the 5DS uses a NEW MODULE, even if it is performing a function they did on APS-C a bit earlier. Even if you think Canon's biggest advance in camera bodies this year is a FF lens (really?), that is no more helpful to their APS-C mirrorless aspirations.

My point was that Canon has not been on a fast track to compete with other mirrorless bodies, and it actually sounds like you're in full agreement about that.

Direct link | Posted on Feb 12, 2015 at 23:34 UTC

I don't think Canon is very afraid. I think they're mostly trapped in a development cycle that keeps most subsystems unchanged across many years, upgrading only one or two at a time. This year, apparently susbsystem they are most proud of is the new mirror drive mechanism... which isn't helping them make a more viable mirrorless product!

Anyway, the EOS M3 is headed in the right direction, but incremental evolution hasn't been fast enough to keep up with Sony or Samsung. No point in releasing the M3 in the US where it would probably hurt sales by effectively blessing the approach better executed by Sony, etc. The lack of native lenses isn't a big problem in reality, because the M models all support adapting EF and EF-S lenses (Canon couldn't re-engineer more than a few per year as native EOS-M mount anyway).

BTW, I have an EOS-M, and overall it is severely outperformed by my Sonys (even the original NEX-5), but I have it because it can be reprogrammed using Magic Lantern (ML).

Direct link | Posted on Feb 12, 2015 at 04:26 UTC as 80th comment | 3 replies
In reply to:

Sdaniella: Whilst Samsung's NX1 is the first affordable consumer 4k APS-C (near Super35mm Cine sized sensor) mirrorless digital camera (great for budding 4k Cinematographers) ... with fast AF

This new Canon EOS 760D serves as the first affordable consumer HDR APS-C (near Super35mm Cine sized sensor) touch-screen capable AF for budding Cinematographers wanting easy-access (minimal or no post processing) for Dual-Level Exposure for HDR CINE = like a beginners version of pro industry ARRI ALEXA CINE cams minus the expense, weight, heat, bulk, and conventional MF lens methods ... brings to the everyday shooter 'normalized' HDR CINE! (@ 720p = managable filesizes for global online sharing)

nice!

it's a preview of what's to come in higher models, like EOS 80D? (HDR 1080p) 5DMkIV? (HDR 4K, HDR 1080p), etc
EOS C-series: 1D or C#00
and likely Powershots, too

and no need to resort to ML (Magic Lantern hack), although I'm sure there will be plenty of new features buried yet to unlock

+lots o' lenses!

Sdaniella, I don't know where you're getting these wild DR claims for Canons, because they disagree with everything I've seen and measured... by a lot. BTW, DxO knows what they are doing and does it very carefully. Basically, Canon has an ancient chunk of their imaging pipeline still in use, and it's really hurting them.

Direct link | Posted on Feb 12, 2015 at 00:05 UTC
In reply to:

Prognathous: Among all weather sealed zoom lenses for APS-C DSLRs, this one is the shortest. Remove this elaborate qualification, and it's just another compact zoom :-)

Weather resistant is NOT a phase that mixes well with collapsible. That's rather special and kinda scary if I'm being honest. 9 seals sounds impressive, but to me that just says "there were nine holes we figured we better plug" -- it's quite possible to design a lens such that the number would be more like 3. I'd hold off on applause until we see an official ratng of how weather resistant this really is....

Direct link | Posted on Feb 11, 2015 at 05:02 UTC
On Sigma goes wide with 24mm F1.4 DG HSM Art lens article (184 comments in total)

"SIgma goes wide" is a an anachronistic understatement. Sigma has been ruling the ultrawide lens market for some time with things like their 8-16mm and 12-24mm zooms, and 24mm isn't very wide for them, nor is it even the widest of their Art line so far (tied for 3rd place).

Direct link | Posted on Feb 10, 2015 at 21:51 UTC as 10th comment | 1 reply
In reply to:

jtan163: I've seen a lot of people say that the 5ds used the 7d2 sensor tech.
However Chuck Westfall did not say that.
He said it used the same pixel size as and the same processor (i.e. DIGIC version).

He also said the same DR as the 5d3 and noise about the same as the 7d2.
To me that does not imply the same sensor tech as the 7d2.

If the exact same sensel design and fab are used, yield on the FF sensor will be pretty terrible, bumping sensor cost an order of magnitude. My guess would be that Canon is in fact doing that, but increasing yield by setting a lower threshold for acceptable performance... which would explain the lower max ISO. It wouldn't shock me if max ISO went up as they got more experience tweaking the fab line.

Direct link | Posted on Feb 8, 2015 at 20:44 UTC
In reply to:

Sdaniella: Whilst Samsung's NX1 is the first affordable consumer 4k APS-C (near Super35mm Cine sized sensor) mirrorless digital camera (great for budding 4k Cinematographers) ... with fast AF

This new Canon EOS 760D serves as the first affordable consumer HDR APS-C (near Super35mm Cine sized sensor) touch-screen capable AF for budding Cinematographers wanting easy-access (minimal or no post processing) for Dual-Level Exposure for HDR CINE = like a beginners version of pro industry ARRI ALEXA CINE cams minus the expense, weight, heat, bulk, and conventional MF lens methods ... brings to the everyday shooter 'normalized' HDR CINE! (@ 720p = managable filesizes for global online sharing)

nice!

it's a preview of what's to come in higher models, like EOS 80D? (HDR 1080p) 5DMkIV? (HDR 4K, HDR 1080p), etc
EOS C-series: 1D or C#00
and likely Powershots, too

and no need to resort to ML (Magic Lantern hack), although I'm sure there will be plenty of new features buried yet to unlock

+lots o' lenses!

Honestly, I'd be getting a 5DS too... except I'm pretty sure there will be a better option from Sony by the time the 5DS is really shipping. I do like the fact that 50MP FF is above Nyquist for just about every lens, including on APS-C crops.

Direct link | Posted on Feb 8, 2015 at 14:58 UTC
In reply to:

Sdaniella: Whilst Samsung's NX1 is the first affordable consumer 4k APS-C (near Super35mm Cine sized sensor) mirrorless digital camera (great for budding 4k Cinematographers) ... with fast AF

This new Canon EOS 760D serves as the first affordable consumer HDR APS-C (near Super35mm Cine sized sensor) touch-screen capable AF for budding Cinematographers wanting easy-access (minimal or no post processing) for Dual-Level Exposure for HDR CINE = like a beginners version of pro industry ARRI ALEXA CINE cams minus the expense, weight, heat, bulk, and conventional MF lens methods ... brings to the everyday shooter 'normalized' HDR CINE! (@ 720p = managable filesizes for global online sharing)

nice!

it's a preview of what's to come in higher models, like EOS 80D? (HDR 1080p) 5DMkIV? (HDR 4K, HDR 1080p), etc
EOS C-series: 1D or C#00
and likely Powershots, too

and no need to resort to ML (Magic Lantern hack), although I'm sure there will be plenty of new features buried yet to unlock

+lots o' lenses!

Dr_Jon, the Canon EOS sensors basically have about 1EV less total DR available than the Sonys, but the analog processing in the Canons flattens out DR at low ISOs, clipping where you'd see Sony-like high DR. DxO has the full resolution DR measures posted as graphs, and if you hover over a point, it will tell you the precise value (how I got the numbers I posted above). Unfortunately, when you go through calibrated combining of everything, it really does come out to only ~3EV gain possible by combining different ISOs even using every ISO the camera supports.That's still great... but Sony has been making APS-C sensors that directly output such great DR for several years.

My guess is that Canon is still using an ancient pipeline to process sensor samples. Actually, I kind of know they are, because CHDK and ML expose some of it. Anyway, Canon needs to stop doing that. ;-)

Direct link | Posted on Feb 8, 2015 at 14:21 UTC
In reply to:

Sdaniella: Whilst Samsung's NX1 is the first affordable consumer 4k APS-C (near Super35mm Cine sized sensor) mirrorless digital camera (great for budding 4k Cinematographers) ... with fast AF

This new Canon EOS 760D serves as the first affordable consumer HDR APS-C (near Super35mm Cine sized sensor) touch-screen capable AF for budding Cinematographers wanting easy-access (minimal or no post processing) for Dual-Level Exposure for HDR CINE = like a beginners version of pro industry ARRI ALEXA CINE cams minus the expense, weight, heat, bulk, and conventional MF lens methods ... brings to the everyday shooter 'normalized' HDR CINE! (@ 720p = managable filesizes for global online sharing)

nice!

it's a preview of what's to come in higher models, like EOS 80D? (HDR 1080p) 5DMkIV? (HDR 4K, HDR 1080p), etc
EOS C-series: 1D or C#00
and likely Powershots, too

and no need to resort to ML (Magic Lantern hack), although I'm sure there will be plenty of new features buried yet to unlock

+lots o' lenses!

PS: Video is even worse than stills. Although you might think the lower resolution would help, and it does, electronic shuttering reduces SNR enough to essentially cancel that advantage, and then there are the lossy video encoding issues. The extra processing of video makes it difficult to get as precise measurements of DR, but it clearly is never much better than stills and is often far worse. In fact, DR unexpectedly goes down on most cameras in the continuous shooting modes vs. one frame/shutter button press mode... perhaps due to more electromagnetic noise from digital processing overlapping analog processing?

Direct link | Posted on Feb 8, 2015 at 12:07 UTC
Total: 696, showing: 141 – 160
« First‹ Previous678910Next ›Last »