What tech advancements do you think will eventually come to prosumer video?

[kye]

2 hours ago, webrunner5 said:

Price is not too bad either. 349 bucks.

https://www.adorama.com/wrllite.html

Jeez.

This would have to be one of the best production-value-per-dollar things you can buy.  In the list under "buy first camera", "buy first microphone", "buy first tripod", and "learn how to use lighting"

[tellure]

10 hours ago, androidlad said:

One ADC per pixel is called Electronic Global Shutter.

For us laypersons can you elaborate on what benefits this brings?  Does this get us per-pixel HDR?

[ND64]

3 minutes ago, tellure said:

For us laypersons can you elaborate on what benefits this brings?  Does this get us per-pixel HDR?

You get lower DR at high ISO cause the amplification needed to increase the brightness of darker areas is too much for already bright areas. With this new pixel design, it could be possible to amplify only those pixels that need to be amplified. So you get near ISO 100 DR with very low noise at say ISO 12800.

Other than fabrication complexities, there is big problem in profiling the output, because it's not going to be linear, I guess. 

[tellure]

36 minutes ago, Eric Calabros said:

You get lower DR at high ISO cause the amplification needed to increase the brightness of darker areas is too much for already bright areas. With this new pixel design, it could be possible to amplify only those pixels that need to be amplified. So you get near ISO 100 DR with very low noise at say ISO 12800.

Other than fabrication complexities, there is big problem in profiling the output, because it's not going to be linear, I guess. 

Ah, thanks.  That does sound pretty awesome.

[kye]

2 hours ago, Eric Calabros said:

You get lower DR at high ISO cause the amplification needed to increase the brightness of darker areas is too much for already bright areas. With this new pixel design, it could be possible to amplify only those pixels that need to be amplified. So you get near ISO 100 DR with very low noise at say ISO 12800.

Other than fabrication complexities, there is big problem in profiling the output, because it's not going to be linear, I guess. 

There might be a clever way to keep it aligned, for example if the gain on each pixel only had a few settings to choose from, but each was 2X or 4X the last.  Then you read the value of the pixel and only need to do a bit-shift to the value to align it with the rest of the image.  Audio DACs are 16 or 24-bit, and each bit is worth double the previous one, so it wouldn't be that challenging to also make the amplifier for each one have a few "bits", and I'm sure there are some simple circuits that can use the amplifier setting to shift the bits in what the ADC is putting out.

 

[fuzzynormal]

"What tech advancements do you think will eventually come to prosumer video?"

All of them. In 15 years the IQ tech will be so good, mundane, and readily available that the camera will just be a minor thought when it comes to production. 

[webrunner5]

22 hours ago, tellure said:

For us laypersons can you elaborate on what benefits this brings?  Does this get us per-pixel HDR?

Pretty much eliminates Rolling Shutter and helps with Banding. Two of the big gains.

[webrunner5]

I don't know a lot about HDR in video. But I am sure it has to be 10 bit or better and have a Log profile to do it. So no clue what per-pixel ADC helps on that part for HDR?

 

[tellure]

11 minutes ago, webrunner5 said:

I don't know a lot about HDR in video. But I am sure it has to be 10 bit or better and have a Log profile to do it. So no clue what per-pixel ADC helps on that part for HDR?

 

Just theorizing here since I don't know a ton about sensor design but if you have a per-pixel ADC then could you, for example, expose for the blown-out windows in an indoor shot and then boost the gain only on the darker pixels to get the equivalent of a bracketed HDR shot?

[kye]

14 hours ago, tellure said:

Just theorizing here since I don't know a ton about sensor design but if you have a per-pixel ADC then could you, for example, expose for the blown-out windows in an indoor shot and then boost the gain only on the darker pixels to get the equivalent of a bracketed HDR shot?

I think that's the idea.  

If I understand it right, some pixels could be at ISO 100 and others at ISO 25000, so your DR would go through the roof.  However, if you had a part of the image that was very dark and those pixels had high ISO, then they'd still be noisy.

[androidlad]

15 hours ago, tellure said:

Just theorizing here since I don't know a ton about sensor design but if you have a per-pixel ADC then could you, for example, expose for the blown-out windows in an indoor shot and then boost the gain only on the darker pixels to get the equivalent of a bracketed HDR shot?

 

26 minutes ago, kye said:

I think that's the idea.  

If I understand it right, some pixels could be at ISO 100 and others at ISO 25000, so your DR would go through the roof.  However, if you had a part of the image that was very dark and those pixels had high ISO, then they'd still be noisy.

Sensors cannot distinguish image content and adjust exposure on a per area/per subject basis.

Sony has a working 1/1.55" automobile sensor IMX490 that achieves 20 stops (120dB) dynamic range in a single exposure. Basically a 3um photosite with a 0.9um subpixel. Three readout for every exposure - big pixel + high gain, big pixel + low gain and small pixel, them merged to get an 120dB signal.

[kye]

2 hours ago, androidlad said:

 

Sensors cannot distinguish image content and adjust exposure on a per area/per subject basis.

Sony has a working 1/1.55" automobile sensor IMX490 that achieves 20 stops (120dB) dynamic range in a single exposure. Basically a 3um photosite with a 0.9um subpixel. Three readout for every exposure - big pixel + high gain, big pixel + low gain and small pixel, them merged to get an 120dB signal.

20dB would be just fantastic!  What bit depths are they talking?  No point having 20 stops of DR and then only having 10 or 12 bits - by the time you convert back to a standard gamma curve it would be banding central.

Could you post a link to your source?

[webrunner5]

Only 10 or 12 bits. A little bit better than 2K output.

 

"Key Specifications

Model nameIMX490

Number of effective pixels2896 (H) × 1876 (V), approx. 5.4 megapixels

Image sizeDiagonal 10.36mm (type 1/1.55)

Unit cell size3.0μm (H) × 3.0μm (V)

Frame rateFull pixelAD10bit 40fps, AD12bit 30fps

Sensitivity (F5.6 standard value, 1/30 second storage time)2,280mV (green pixel)

Dynamic range (EMVA 1288 standard)120dB (140dB when prioritizing dynamic range)

Power supplyAnalog3.3V

Digital1.8V1.1V

Interface1.1V1.8V

InterfaceMIPI CSI-2 serial output (4 lane / 2 lane)

Package124pin plastic BGA

Package size15.35mm x 11.68mm"

 

Sensor size. 1/1.5. Not too bad. Smartphone size I guess.

 

 

[kye]

7 hours ago, webrunner5 said:

Only 10 or 12 bits. A little bit better than 2K output.

Sensor size. 1/1.5. Not too bad. Smartphone size I guess.

Actually, done a bit more reading about how log profiles use bit-depth and 12-bit would be fine for 20 stops of DR.  If it was only 10-bit then there would be less bits-per-stop than other 10-bit profiles, although they could probably re-arrange how many bits each stop gets and get away with it - the extreme highlights and shadows wouldn't need as many bits.

 

[androidlad]

8 hours ago, webrunner5 said:

Only 10 or 12 bits. A little bit better than 2K output.

 

"Key Specifications

Model nameIMX490

Number of effective pixels2896 (H) × 1876 (V), approx. 5.4 megapixels

Image sizeDiagonal 10.36mm (type 1/1.55)

Unit cell size3.0μm (H) × 3.0μm (V)

Frame rateFull pixelAD10bit 40fps, AD12bit 30fps

Sensitivity (F5.6 standard value, 1/30 second storage time)2,280mV (green pixel)

Dynamic range (EMVA 1288 standard)120dB (140dB when prioritizing dynamic range)

Power supplyAnalog3.3V

Digital1.8V1.1V

Interface1.1V1.8V

InterfaceMIPI CSI-2 serial output (4 lane / 2 lane)

Package124pin plastic BGA

Package size15.35mm x 11.68mm"

 

Sensor size. 1/1.5. Not too bad. Smartphone size I guess.

Remember it's 3 readout for a single exposure, even if the ADCs only operate at 10bit, it would be equivalent of a 30bit ADC.

[kye]

6 minutes ago, androidlad said:

Remember it's 3 readout for a single exposure, even if the ADCs only operate at 10bit, it would be equivalent of a 30bit ADC.

Absolutely.  It would be a wonderful signal coming off the sensor.

I was thinking more about what the output file would be, as if you have the same bit-depth but greater DR than the existing gamma curve is designed to handle then you need to compress more DR into the same number of bits and you risk the banding problems that 8-bit Log can suffer from.  But after doing a bit more reading I figure they could probably get away with it without huge issues, even if it was still 10-bit.