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 and creatively using camera systems to provide new abilities and improved quality.

Comments

Total: 1708, showing: 61 – 80
« First‹ Previous23456Next ›Last »
On article Sony 'striping': here's the fix (791 comments in total)
In reply to:

Vincent DP: Look at the hairs that cover her right eyelid... they are messed up by the workaround.
Anyway, better than no workaround at all.

Blonde hair strands in the right orientation look enough like the artifact to confuse the current version of KARWY-SR. When I bring KARWY-SR out of beta, it will probably be better at distinguishing. However, if we apply the current logic with additional knowledge of PDAF cell placement for that camera model, such errors become extremely unlikely -- that's what Sony should consider putting in camera.

Link | Posted on Apr 18, 2018 at 17:15 UTC

Large format digital will be a big thing... no pun intended. The catch is that these first sensors don't have the high resolution that will make it compelling for new uses, but the images on their site are indeed pretty compelling anyway -- pretty much for the reasons Rishi explained. Only thing here that's news to me is the price... which is high, but not insanely high.

Link | Posted on Apr 18, 2018 at 00:52 UTC as 78th comment
In reply to:

ProfHankD: Just to be clear, this appears to be a conventional DOF adapter, which basically means it is a camera obscura: the image is projected on a "film" plane and then that film plane is imaged by the cell phone lens. The "film" plane would be a material that acts as a focus plane, which essentially means a diffuser, such as ground glass or translucent plastic. The key problem is that the diffuser's texture will show, so most serious rigs incorporate a mechanism that physically moves/vibrates the diffuser so that any texture is smoothed; many use rotating ground glass, but vibrating would seem more likely here.

Without the motion of the diffuser, these are trivially home-made. Perhaps I'll throw-together a 3D-printed one just for laughs...?

Here's my old 3D-printed digital pinhole obscura: https://www.dpreview.com/forums/post/57666254

Looking at the video again, I doubt they're vibrating the diffuser....

Link | Posted on Apr 13, 2018 at 01:00 UTC

Just to be clear, this appears to be a conventional DOF adapter, which basically means it is a camera obscura: the image is projected on a "film" plane and then that film plane is imaged by the cell phone lens. The "film" plane would be a material that acts as a focus plane, which essentially means a diffuser, such as ground glass or translucent plastic. The key problem is that the diffuser's texture will show, so most serious rigs incorporate a mechanism that physically moves/vibrates the diffuser so that any texture is smoothed; many use rotating ground glass, but vibrating would seem more likely here.

Without the motion of the diffuser, these are trivially home-made. Perhaps I'll throw-together a 3D-printed one just for laughs...?

Here's my old 3D-printed digital pinhole obscura: https://www.dpreview.com/forums/post/57666254

Link | Posted on Apr 13, 2018 at 00:40 UTC as 4th comment | 2 replies

Basically, it sounds like Photokina just cut the benefit to exhibitors in half while effectively tripling exhibitor cost per unit time.

The cost of fielding an exhibit is pretty high: http://www.photokina.com/photokina/For-Exhibitors/Application/Calculate-stand-costs/index.php

Odd move for Photokina at a time when the industry seems troubled. Almost feels like a desperate attempt to draw more attention to the field...? If so, I hope that works.

Link | Posted on Apr 10, 2018 at 16:11 UTC as 32nd comment | 2 replies
In reply to:

A Owens: Dang, that tripod pixel shift looks good.

A Owens: Sony's NEX line did computational multi-shot processing to make-up for lack of IBIS, and Sony was pretty aggressive about it. It sure looks to me like they implemented sub-pixel alignment with quality-based region rejection and even warping to allow for lens distortion in movements of many pixels. They also didn't just use it for one thing, but for anti-shake, high-ISO noise reduction, 2D/3D panorama stitching, etc. Anyway, spatial SR and related algorithms were well known by the early 2000s -- they were just too computationally expensive for most cameras to implement, but it looks like Sony implemented them with hardware assist in the JPEG pipe (which would explain why no raw output).

Pentax seems to be using much simpler algorithms for PS. Artifacts include line and cross-hatch textures on some of the details, especially from the hand-held PS shots. For example, look at image #2 above the door where it says "IN SEMINE".

Link | Posted on Apr 10, 2018 at 00:54 UTC
In reply to:

A Owens: Dang, that tripod pixel shift looks good.

Actually, the PS shots all look a bit choppy. The straight shots look excellent. :-)

It's rather difficult to ensure pixel-level alignment because camera, lens, and scene components often move a bit -- as I found several years ago, even with a camera bolted to a concrete block, lens elements can shift within play of the barrel.

As far back as the NEX-5, Sony has been doing computational multi-shot alignment and enhancement (e.g., for anti-shake and noise reduction), and I don't think this is really doing any better and maybe not quite as well. The catch is that Sony's processing only produces a JPEG result: no raws.

Link | Posted on Apr 9, 2018 at 23:26 UTC
In reply to:

ProfHankD: "By then [before 2020], it would make sense for both Canon and Nikon to have replicated their most important tele primes and wide-aperture zoom lenses in their respective new native mirrorless mounts"

That's highly unlikely. Releasing adapters to whatever new mount (like Sony) is much more likely, but Sony is now years ahead ON NEW FF MIRRORLESS LENSES. Pentax tried the other way forward -- mirrorless using the SLR mount -- and it wasn't a resounding success. Canon's EF-M mount could be used for a FF mirrorless, so they're actually a step ahead of Nikon if they decide to go that route. Fuji's "oversized FF" actually seems a smarter upgrade path.

The MFT makers have a bigger problem: the 4:3 aspect choice was wrong, the guess that bigger sensors would be prohibitively expensive was wrong, and they're stuck with that choice. They even have a relatively long flange distance....

cdembrey: Nice try, but the logic behind FourThirds was primarily CRTs for PCs and TVs, and MFT inherited that already-outdated decision. Aspect ratio cropping wouldn't matter so much if MFT used a bigger sensor, but cropping 4:3 to match the now-dominant 16:9 standard, or even 3:2, really hurts. Resolution expectations have risen beyond what's easy for lenses to deliver on an MFT sensor; new MFT lenses often meet the challenge, but many older lenses don't.

The big problem for MFT is that the entire FourThirds system was designed with very little wiggle room for upgrading. Sony's E mount just barely allowed for FF use, but you can bet that's not a coincidence; Canon's EF-M leaves even more margin for FF. MFT is basically stuck without an easy upgrade path. You can argue it's good enough, and in many ways it is, but it's going to get harder and harder for MFT to be compelling against cell phones; I think MFT's future is in space/weight constrained drones, etc.

Link | Posted on Apr 9, 2018 at 02:35 UTC

"By then [before 2020], it would make sense for both Canon and Nikon to have replicated their most important tele primes and wide-aperture zoom lenses in their respective new native mirrorless mounts"

That's highly unlikely. Releasing adapters to whatever new mount (like Sony) is much more likely, but Sony is now years ahead ON NEW FF MIRRORLESS LENSES. Pentax tried the other way forward -- mirrorless using the SLR mount -- and it wasn't a resounding success. Canon's EF-M mount could be used for a FF mirrorless, so they're actually a step ahead of Nikon if they decide to go that route. Fuji's "oversized FF" actually seems a smarter upgrade path.

The MFT makers have a bigger problem: the 4:3 aspect choice was wrong, the guess that bigger sensors would be prohibitively expensive was wrong, and they're stuck with that choice. They even have a relatively long flange distance....

Link | Posted on Apr 9, 2018 at 01:15 UTC as 184th comment | 6 replies
In reply to:

ProfHankD: Are they really selling a semi-fisheye for $700?
Good converter optics (e.g., Raynox) for that cost well under $100.

Trebor1: What? Yeah, the base has 3 elements... I don't see anything exciting about that. Basically, this cuts the production cost by sharing the last few elements, focus, and aperture, and it does make the front parts of the lens smaller. However, they don't seem to be passing-on the cost savings, the size advantage only happens if you get multiple front lenses, flare seems pretty bad, and even the rather favorable review you cited basically has this not very sharp on 16MP FF (which suggests it would struggle to produce sharp 6MP APS-C).

I fully understand that paying several times the price to get a smaller, funkier, optic appeals to some hipsters, but I see very little to recommend this.

Link | Posted on Apr 7, 2018 at 02:08 UTC

Are they really selling a semi-fisheye for $700?
Good converter optics (e.g., Raynox) for that cost well under $100.

Link | Posted on Apr 6, 2018 at 02:32 UTC as 14th comment | 4 replies
On challenge No lens (1 comment in total)

There was an image submitted from a scanner. I was uncertain about this image and there has now been a well-phrased complaint to the effect that, in fact, the scanner used does use a lens to focus the image. That is correct, so I have disqualified the questionable submission.

Link | Posted on Apr 5, 2018 at 11:57 UTC as 1st comment
On article Quick look: Canon's new compressed Raw format (241 comments in total)
In reply to:

ProfHankD: Hmm. Noise frequency changes pretty dramatically... smells a bit like JPEG DCT and YUV encoding are being used. The difference between the images also is not consistently small, which suggests some log encoding. Overall, I'm not sure this is buying anything you couldn't get by saving JPEGs using one of the log-based encoding schemes now so popular for extending video DR... is it?

The block artifacts are from DCT on 8x8 pixel blocks, and this compressed raw format's noise texture looks like it may have gone through the same DCT. JPEG2000 is better, and although it is 2018, I don't know of any consumer camera that has moved from the circa-1996 JPEG encoding to that. Some compressed raw formats basically use the JPEG compression engine with a few tweaks or adding-in an also-compressed error image, and without knowing what Canon is doing in this format, something along those lines would be my guess....

Link | Posted on Apr 3, 2018 at 12:15 UTC
On article Quick look: Canon's new compressed Raw format (241 comments in total)

Hmm. Noise frequency changes pretty dramatically... smells a bit like JPEG DCT and YUV encoding are being used. The difference between the images also is not consistently small, which suggests some log encoding. Overall, I'm not sure this is buying anything you couldn't get by saving JPEGs using one of the log-based encoding schemes now so popular for extending video DR... is it?

Link | Posted on Apr 2, 2018 at 15:17 UTC as 54th comment | 3 replies

"same image circle size as a full frame DSLR,* meaning the C700 FF can use Canon's entire line of EF-mount lenses at their standard focal lengths"

That's terrible rephrasing of a claim Canon also phrased badly:

1. The focal length of a lens is not altered by the size or shape of the sensor behind it. If you meant equivalent focal length, the wider sensor actually increases field of view a little compared to 16:9 from a 3:2 FF sensor.

2. The image circle isn't actually circular for some of Canon's EF lenses: some have hoods or contain internal masks approximating the 3:2 ratio. I would expect most of the problematic lens parts could be removed or modified easily enough.

Otherwise, a good step forward for Canon... although I'm not likely to buy one. ;-)

Link | Posted on Mar 28, 2018 at 14:56 UTC as 40th comment

Really?

I don't think pre-flash-exposed == pre-colored, although pre-flashing can slightly improve tonal range (an old trick against reciprocity failure). Monsoon looks fairly normal and predictable, which I suspect is why it's the big seller. I don't know who'd want a Sunstroke.

Link | Posted on Mar 27, 2018 at 15:09 UTC as 40th comment

Ok, so we're looking at Cinema lenses and measuring Resolution. Is Resolution really a big concern for video?

Well, APS-C is 23.7mm wide and 1080 video needs 1920 pixels, so that's 12.3um per pixel or one line pair per 24.6um -- which is 41 lp/mm. MFT is 17.30mm wide and 4K needs 3840 pixels, so that's 4.5um per pixel or one line pair per 9um -- which is 111 lp/mm. Although these lenses will not cover FF, you would still need about 27 lp/mm to be pixel-level sharp for 1080. Note that I'm not saying current sensors have enough pixels to get Nyquist samplings at those resolutions, because most are not even close, but ideal video lens Resolution is surprisingly high.

In sum, very impressive performance and it might matter. That said, I still can't quite see calling a $1000 lens cheap even if it competes well with $30,000 lenses. I suppose it might be cheap enough to cut into rentals of the more expensive alternatives.... ;-)

Link | Posted on Mar 27, 2018 at 14:24 UTC as 6th comment
In reply to:

makofoto: We use a LOT of GoPros in the movie industry. I've used as many as 27on a NISSAN commercial. The reason GoPro's are worth the money to us is they offer ProTune, a low contrast, double sized file ... so we can match better with our Arri Alexa main cameras ... and they are reliably good and sharp. For years I've tested everything new that comes out. So often their lenses would not be sharp across the frame. Every GoPro has been fine. But yeah, for kids, I would recommend Chinese. Typical pile of gear: https://images14.fotki.com/v699/fileXmmG/caef1/4/43793/4909192/image6.jpg

makofoto: In my research, I use fleets of Canon PowerShots reprogrammed using CHDK. My latest fleet is a couple dozen SX530 HS: 16MP with rectilinear 50X zoom cameras that cost me $130 each as factory refurbs. They aren't particularly rugged nor waterproof, but they are very capable cameras. I would have expected them, or the new little Sonys, to be be more useful tools for cinematography.

I can see the appeal of ProTune, but does that really make-up for the semi-fisheye lens, lack of programming support, etc.?

Link | Posted on Mar 27, 2018 at 00:05 UTC
In reply to:

ProfHankD: There's been research and active discussion in the forums trying to fully understand the strange flare-related stripe artifacts that showed-up in some of the motorcycle couple shots. I see a few new shots here that tried to get that defect, but didn't. The quick summary is that the defect is rare shooting scenes with a lens attached, although it's easy enough to get with a bright light and no lens. Beyond that, RawTherapee now has a patch that can remove the defect and I have posted KARWY-SR, http://aggregate.org/DIT/KARWY-SR/ , free software that runs in your WWW browser and produces credibly-repaired compressed ARW2 raw files.

In sum, I think the primary weak point of this camera -- the stripe defect which also can occur with some other Sony models -- is basically a non-issue. It was just a very unfortunate set of circumstances that made the defect visible in some of the first sample images out there.

Magnar W: That's too crude for stripes on a textured area, such as the images where the model's face and hair are affected. In many cases, something as simple as the GIMP despeckle or destripe filters can suffice.

Unfortunately, the repairs really need to be done on raw data. Interpolation and lens-correction transformations spread the defect in ways that are not fully repairable.

RawTherapee uses a table of PDAF positions and a simple pattern recognizer to identify pixels to repair; KARWY-SR uses a slightly fancier pattern recognizer so it doesn't need a PDAF position table. RawTherapee repairs using its "bad pixel" interpolation logic; KARWY-SR repairs using very crude texture synthesis with shaped noise. Not surprisingly, RawTherapee and KARWY-SR both make better-quality repairs than is feasible with Photoshop... and KARWY-SR actually produces a modified compressed ARW2 raw file so you can use whatever postprocessing tools after that and not have problems.

Link | Posted on Mar 25, 2018 at 17:56 UTC
In reply to:

keeponkeepingon: Wow. 171 images and not a single one taken with the lens sony sells with the camera? (28-70 mm )

While it is nice to see these pictures with the pricey sony glass (other than the few with the $550 85mm prime, every other lens is $800-$1700), It'd be really nice to get a few pictures with the kit lens to give folks an idea if they should spend the extra $200 for the 28-70.

Actually, the kit zoom looks more than decent as a no-brainer one-lens choice. Dyxum reviewers like it 4.09/5 overall: http://www.dyxum.com/lenses/Sony-FE-28-70mm-F3.5%E2%80%935.6-OSS_lens724.html -- the general conclusion is that it's fairly well-made and optically good in the center, but the edges not so much.

Personally, most of the lenses I use are old manual lenses with average cost well under $100 (actually, my 200+ lenses averaged about $25 each)... and they can even autofocus using the TechArt Pro LM-EA7. Even adding the LM-EA7 cost, I'd still have a bag full of lenses for a total cost of $550. This is the real advantage of mirrorless: to make effective use of lenses in any mount. That said, I also have a $1500 Sony STF 100mm -- because there is no other lens that can do what it can do: perfectly apodized bokeh with fast autofocus.

Link | Posted on Mar 25, 2018 at 16:14 UTC
Total: 1708, showing: 61 – 80
« First‹ Previous23456Next ›Last »