Sony Sees Next Big Hit at 1,000 Photos a Second | Hacker News Reader

Sony Sees Next Big Hit at 1,000 Photos a Second

bloomberg.com

30 pointscm218730 comments

Isn't one of the big challenges of hi-speed video how to collect enough photons in such a short time? Super-bright lights are often used to illuminate the subject

'Shooting at a frame rate of 1000fps requires 5.25 times more light than at 24 or 25 fps' [http://www.lovehighspeed.com/lighting-for-high-speed/]

They've been making some headway into low-light performance for their sensors, which undoubtedly helps here.

low light performance is the (one of) holy grail of all camera manufacturers, and one of prime properties evaluated when new hardware arrives (be it lens or sensor). don't expect any major breakthrough there, low hanging fruit has been covered looong time ago.

Still the current generation of Sony A7 cameras produce some pretty damn usable results at both 6400 and 12800 ISO, allowing 1/1000 - 1/2000 second shutter time in 'normal' lighting conditions.

a7s goes up to 409600 ISO, and some would say the results at 80000 ISO are pretty damn usable [http://philipbloom.net/blog/a7s/]!

Canon released an ISO 4000000 camera recently

http://petapixel.com/2015/07/30/canon-just-unleashed-an-iso-...

AFAIK the problem with the old sensors was that they were manufactured at 100s of nm which was really bad for low light. As Sony is now pushing sensor manufacturing down to CPU-level, the sensitivity increases significantly. Once they hit the current cutting edge 14nm process, we can expect significant low-light gains as well.

Isn't it exactly the opposite? The smaller the area, the less photons land on it in the given time, so the sensors with the bigger pixels can capture more light, and the less nm features measure the more of noise and less of signal.

That's why if you want good pictures in low light you need a camera with as big sensor as you can get.

http://www.dpreview.com/articles/5496399487/canon-multi-purp...

"At the core of the ME20F-SH is a 2.26 megapixel CMOS sensor, originally announced in 2013, which has pixels measuring 19μm - 5.5X larger than what's found on high-end DSLRs. This allows for 1080/60p/30p/24p (and PAL equivalent) video capture in light levels as low as 0.0005 lux at a maximum gain setting of 75 Db, which is equivalent to over ISO 4,000,000."

Note: the pixel is larger and therefore the sensor more sensitive. Canon made the full frame sensor with only 2 megapixels and obtained the higher sensitivity.

Comparing with the cameras in phones, the area of the whole sensor in iPhone 6 is 50 times smaller than the area of the full frame sensor.

Making the components smaller doesn't mean you have to make the sensor area smaller, you can just put more of them in.

> Making the components smaller doesn't mean you have to make the sensor area smaller, you can just put more of them in.

It sounds, mildly put, different to all I know, so much that I fully doubt the validity of claim, as long as the subject are the camera sensors. My argument is exactly that making the transistors smaller benefits the CPUs but making the pixels smaller makes the sensors worse. The less pixels in the same area, the more sensitivity. So what do you mean can be "put more of" "in"?

I've also linked to an example. The Canon's camera has one pixel of 19μm. The latest Intel CPUs use lithography of 14 nm, that's more than 1000 times smaller in one dimension. Which means that the difference in useful area gained by reducing the frame of the fixed-area pixel as the effect of the lower-nm lithography brings extremely little (for example, twice as thin frame (e.g. 28 nm to 14 nm step) brings just 0.2% more photons in the same area).

Do you have any link?

I think the important distinction you are missing is that smaller manufacturing process does not necessarily imply smaller photosites.

No, improved lithography certainly can reduce the area that is not collecting the photons, I argue that the changes in lithography (nanomenters) are orders of magnitude smaller than the size needed for very high sensitivity (micrometers). I gave the links, and to make this discussion useful, I'd appreciate the links that support different conclusions.

How about smaller, lower-voltage electronic parts cause lower interference, noise and banding, contributing to significant improvement of quality of reading of a sub-pixel value (or pixel layer in Foveon), improving quantum efficiency of the sensor, especially its incident photon to electron ratio? And allowing substrates with better characteristics at smaller scales?

Yes, if you make the pixels smaller, the sensor will get less light sensitive. But for most applications, the sensor size is given (e.g. 35mm cameras) and smaller process sizes would mean, that more of the per-pixel area is taken for light recording and less for the electronic elements on each pixel.

Looking at the sensor under microscope the amount of photons received is determined more by the area of the pixel than the size of the features under that area:

http://petapixel.com/2013/02/12/what-a-dslrs-cmos-sensor-loo...

I'd still like to get a link in support of the claim of significant benefit of lithography process improvement to the sensor sensitivity, unless it's about the smallest sensors (as in mobile phones or smaller) and very high pixel count.

From what I've watched about CMOS sensors and I could be mistaken, the comparison could be similar to making your CPU registers smaller but ramping the frequency of how often you access hit them (which is akin to increased framerate). So the net amount of captured light is the same, but you have many more frames now.

With sophisticated software (or onboard hardware) this can be reassembled to the same picture as well. And you are looking at this a bit wrong way. On mobile devices - your bottleneck is lens size, so you have to have smaller pixels.

Now if your pixels are large, the register (I think they are called wells on CMOS sensors) has to pull in much more information at higher power requirements which introduces noise which you have to "mask" with even more photon energy? Again, I could be getting this totally wrong and of course it's best to have big lens with lots of pixels.

When a Gopro 4 (that's been on the market for a year now) can do 240 fps at 720p resolution, I think it's a bit misleading to say the current state of the art is 60 fps. And it's fair to assume next year's Gopro 5 will have a significant increase on that again.

Are we sure that the Hero5 isn't coming out this year?

Doesn't look like it is confirmed by GoPro, but the rumor that it's being postponed to 2016 has been widely circulated. If GoPro was releasing the Hero5 a month from today, they'd be idiots not to have a press release out denying that rumour.

What I really want, from go pro or Sony is a global shutter. Rolling shutter needs to die.

AFAIK, you can have global shutter on cameras today if you sacrifice low-light performance and high framerates. You need a CCD sensor instead of a CMOS (generally speaking), like you find e.g. on most cameras in the Canon PowerShot line, which do feature global shutters.

Except the article doesn't say that. It says that the "typical high-end camera shoots about 60 photos per second."

Typical high-end is not exactly state of the art, and the distinction is quite clear.

Is there a less blogspammy version of this? The article mentions a YouTube channel but doesn't give us the URL.

Edit: Did a bunch of digging and found it: https://www.youtube.com/channel/UCaq5dRB_CnZDC-eQcsv7nZg

That's nice, how about 14 bit depth raw images in your pro series cameras though?

I mean, the 12+7 images look fine most of the time, but when you're paying close to $3,000 for a body, there should be no compromises in terms of IQ.

Sony owns image sensors atm. They are probably 3-4 years ahead of the competition. They already have consumer 960fs pocket camera out in the market - http://www.sony.net/Products/di/en-us/products/f5kd/

Is 1000fps the next big thing, though? Probably not. (Never heard of the above pocket camera, for example)

I wonder if there are interesting applications of being able to capture images at this kind of speed at high resolution. For example, using image analysis between frames to get some kind of depth map, based on subtle differences between each frame due to slight hand movements.

I was thinking along the same lines. Another specific example would be single-shot HDR or true post-shot-adjustable exposure. E.g. if the optimum exposure for a conventional camera is 1/60th of a second, then instead capture 10-20 exposures whose total time adds up to 1/60th of a second, and sum various numbers of the exposures in software.

Right now that's not practical because DSLRs don't have the hardware capacity to pull that much data off of the sensor and store it quickly enough.

Seems like the additional data would also be very useful for removing motion blur, automatically removing moving objects from e.g. architectural/landscape shots, etc.

The article talks about these sensors not being used for photo or video but rather as vision sensors for cars, robots, etc

Next big thing? Nah. But it's pretty useful in industrial applications. And videographers would definitely take it. Consumers too probably - a lot of mobile phones do limited slo-mo today

Sony is dominating in many different areas in image sensors - high framerates is just one of them. There's a reason they have 40% of the market

Reply on news.ycombinator.com