Mavic Pro-2 vs Zoom and P4P - Line Skipping + Sensor Heat?

[Former Member]

Very interesting and informative thread.

 

[Category 5]

Very interesting and informative thread.

touche. But until we get a firmware alleviating the issue, or a better pic of the top of that chip there's nothing left to do here but argue about semantics and call each other names. lol

 

[Former Member]

touche. But until we get a firmware alleviating the issue, or a better pic of the top of that chip there's nothing left to do here but argue about semantics and call each other names. lol

Your technical contributions are welcome. There are always going to be spirited debates here. Welcome to mavicpilots! hahaha

 

[Cliff_622]

Digital zoom is taking a finite resolution output, and increasing the pixel size to simulate zoom function at the loss of resolution and detail. What the Mavic does is known as a sensor crop, and it is petty much a standard way of achieving 4k output from a non native 4k sensor. It is really the truest form of 4k that can be achieved by the sensor because the output is 1:1 representation of what the sensor sees. Any other output, such as full field of view, requires an algorithm to convert it to 4k, so the final image quality will always be at the mercy of that algorithm. The truest form of 4k that can be achieved from this sensor is HQ crop mode, as it is implemented. Obviously AA filters and post processing will further affect subjective image quality, regardless of how that image is obtained before that stage.

There is no digital zoom of any kind in the M2P.

Actually, the truest form of 4k is not a 1:1 readout from a RGGB Bayer pattern sensor. The truest 4k would come from 3 RGB channels each at 8.2megapixels (3 RGB channels each with 100% full raster source)

What we have in a Bayer 1:1 RGGB readout is a green channel that is only 50% 4k resolution supplemented by a 25% red and 25% blue 4k resolution channels. This is pretty lousy when you stop and think that your BEST channel, (50%) is nowhere near 4k at all!!!..and your red and blue channels are MUCH worse and barely 1080!!

A Sony-style 6k oversample + scaling engine gives your green channel 100% 4k resolution and supplements that with 50% red and 50% blue channeld....much closer to a "real" or "true" 4k image.

1:1 Bayer sensor 4k images are the absolute rock bottom LOWEST pixel count needed that you can legally call your image "4k" with. Its literally the bottom of the 4k barrel.

Cliff Totten

 

[BlueBay]

Actually, the truest form of 4k is not a 1:1 readout from a RGGB Bayer pattern sensor. The truest 4k would come from 3 RGB channels each at 8.2megapixels (3 RGB channels each with 100% full raster source)

What we have in a Bayer 1:1 RGGB readout is a green channel that is only 50% 4k resolution supplemented by a 25% red and 25% blue 4k resolution channels. This is pretty lousy when you stop and think that your BEST channel, (50%) is nowhere near 4k at all!!!..and your red and blue channels are MUCH worse and barely 1080!!

A Sony-style 6k oversample + scaling engine gives your green channel 100% 4k resolution and supplements that with 50% red and 50% blue channeld....much closer to a "real" or "true" 4k image.

1:1 Bayer sensor 4k images are the absolute rock bottom LOWEST pixel count needed that you can legally call your image "4k" with. Its literally the bottom of the 4k barrel.

Cliff Totten

Brilliant as always! Thank you Cliff.

 

[sar104]

Actually, the truest form of 4k is not a 1:1 readout from a RGGB Bayer pattern sensor. The truest 4k would come from 3 RGB channels each at 8.2megapixels (3 RGB channels each with 100% full raster source)

What we have in a Bayer 1:1 RGGB readout is a green channel that is only 50% 4k resolution supplemented by a 25% red and 25% blue 4k resolution channels. This is pretty lousy when you stop and think that your BEST channel, (50%) is nowhere near 4k at all!!!..and your red and blue channels are MUCH worse and barely 1080!!

A Sony-style 6k oversample + scaling engine gives your green channel 100% 4k resolution and supplements that with 50% red and 50% blue channeld....much closer to a "real" or "true" 4k image.

1:1 Bayer sensor 4k images are the absolute rock bottom LOWEST pixel count needed that you can legally call your image "4k" with. Its literally the bottom of the 4k barrel.

Cliff Totten

Is any color sensor specification defined in that way? It's certainly not the conventional way that color sensor resolution is quoted by any manufacturer that I'm aware of.

 

[dwallersv]

I am willing to put up $100 into a pot to get an M2P and tear it down. Who else is willing to chip in?

You must have money to burn.

I'm as curious as anyone here, but I can certainly think of myriad far more valuable things to spend $100 on.

 

[Cliff_622]

Is any color sensor specification defined in that way? It's certainly not the conventional way that color sensor resolution is quoted by any manufacturer that I'm aware of.

Im sorry but I dont understand your question. Are you asking how the Bayer sensor RGGB checkerboard works? If you have a 1million photo site (aka "1 megapixel") sensor, 50% of them are green, 25% red and 25% are blue. It takes 4 photosites to generate 1 single full color pixel.

This is very much the industry sandard ever since Kodak and Bryce Bayer figured how how to make full color images from an RGGB checkerboard grid.

I just purchased a Sony PXW-Z190. This camera does not use 1 single Bayer sensor. It uses THREE image sensors. Each one captures light in full 100% 4k resolution. There is no deBayer process and no color resolution loss. Each RGB channel is 100% prefect. Single sensor camera can never do this. So,...6k (or more) over sampling is the closest they can get.

Is that what you are asking?

 

[sar104]

Im sorry but I dont understand your question. Are you asking how the Bayer sensor RGGB checkerboard works? If you have a 1million photo site (aka "1 megapixel") sensor, 50% of them are green, 25% red and 25% are blue. It takes 4 photosites to generate 1 single full color pixel.

This is very much the industry sandard ever since Kodak and Bryce Bayer figured how how to make full color images from an RGGB checkerboard grid.

I just purchased a Sony PXW-Z190. This camera does not use 1 single Bayer sensor. It uses THREE image sensors. Each one captures light in full 100% 4k resolution. There is no deBayer process and no color resolution loss. Each RGB channel is 100% prefect. Single sensor camera can never do this. So,...6k (or more) over sampling is the closest they can get.

Is that what you are asking?

No - of course I'm not asking how it works - I was pointing out that very few, if any, Bayer sensor manufacturers quote anything other than the number of total pixels - not the number of red, green or blue photosites. So while your observation is correct, it's not how Bayer sensors specifications are quoted. Yes - if you are using a 3-sensor solution the specification is obviously different.

 

[Cliff_622]

No - of course I'm not asking how it works - I was pointing out that very few, if any, Bayer sensor manufacturers quote anything other than the number of total pixels - not the number of red, green or blue photosites. So while your observation is correct, it's not how Bayer sensors specifications are quoted. Yes - if you are using a 3-sensor solution the specification is obviously different.

Gotcha,....yes, the average camera consumer or drone buyer has no clue how image sensors work....and companies like DJI prefer to keep it that way!!!!

An ignorant customer is the BEST customer! ;- )

 

[sar104]

Gotcha,....yes, the average camera consumer or drone buyer has no clue how image sensors work....and companies like DJI prefer to keep it that way!!!!

An ignorant customer is the BEST customer! ;- )

Agreed. Back to the issue - are you aware of any Bayer sensor cameras that use a full complement of each individual color to create 4k?

 

[Cliff_622]

Agreed. Back to the issue - are you aware of any Bayer sensor cameras that use a full complement of each individual color to create 4k?

Yes, almost every Sony camera in the past 4 years. (APSC, 1inch-type and Full frame) and now Panasonic GH5 and G9. Sony does 5k and 6k collection and assembly, preps that image and scales down to 4k....spectacular! The Panasonic GH5 does a 5k 16x9 sensor radout and assembly using a Sony IMX272 sensor.

Yes,...there are probably 15 or more cameras that do this with Sony EXMOR sensors. The DJI Insipre X7 and X5s cameras use Sony sensors and do this too.

Once Sony figured out how to 5k and 6k oversample, they moved away from crude 1:1 readout FAST.

Cliff Totten

 

[sar104]

Yes, almost every Sony camera in the past 4 years. (APSC, 1inch-type and Full frame) and now Panasonic GH5 and G9. Sony does 5k and 6k collection and assembly, preps that image and scales down to 4k....spectacular! The Panasonic GH5 does a 5k 16x9 sensor radout and assembly using a Sony IMX272 sensor.

Yes,...there are probably 15 or more cameras that do this with Sony EXMOR sensors. The DJI Insipre X7 and X5s cameras use Sony sensors and do this too.

Once Sony figured out how to 5k and 6k oversample, they moved away from crude 1:1 readout FAST.

Cliff Totten

OK - that makes sense.

 

[Category 5]

Foveon’s Remarkable Quest for Realism

Looks like Foveon is still doing their thing. When the first sigma cameras using their chips came out they were like 6mp or something but Sigma marketed them as having the equivalent resolution of a 13mp camera. Problem was That everyone else was doing 13+ MP so the advantage seemed wasted. I don’t know if these chips are fast enough for video, but for 4K/8k video it’d be interesting to see how they could work.

 

[Roamer105]

No - of course I'm not asking how it works - I was pointing out that very few, if any, Bayer sensor manufacturers quote anything other than the number of total pixels - not the number of red, green or blue photosites. So while your observation is correct, it's not how Bayer sensors specifications are quoted. Yes - if you are using a 3-sensor solution the specification is obviously different.

Yes it takes 4 (RGGB) component colors (quad) to create full color pixel (at least in the simplest form since debayering algorithms use overlapping RGGB quads they can actually do better and reduce noise that way too). However each component color produces a discrete luminance value that if we ignored colors would produce a 4K monochrome image (luminance only). Even though sophisticated debayering algorithms will produce more color resolution than 1/4, we are recording 4-2-0. Only luminance information is recorded at 4K while color information is recorded only 1/4 of that resolution.

So it is perfectly valid and accurate to call this sensor 4K with 4-2-0 color encoding. Three sensors do not produce more luminance resolution, but they will produce full 4-4-4 resolution color as well.

 

[Cliff_622]

Yes it takes 4 (RGGB) component colors (quad) to create full color pixel (at least in the simplest form since debayering algorithms use overlapping RGGB quads they can actually do better and reduce noise that way too). However each component color produces a discrete luminance value that if we ignored colors would produce a 4K monochrome image (luminance only). Even though sophisticated debayering algorithms will produce more color resolution than 1/4, we are recording 4-2-0. Only luminance information is recorded at 4K while color information is recorded only 1/4 of that resolution.

So it is perfectly valid and accurate to call this sensor 4K with 4-2-0 color encoding. Three sensors do not produce more luminance resolution, but they will produce full 4-4-4 resolution color as well.

Yes, it is valid to call this a "4k" sensor. (Both, the Zoom and the Pro in crop HQ) However, I'm only saying that a 6k (100%+50%+50% 4k) sensor readout can produce a nocicably higher frequency detail. Even using the same 4:2:0 chroma subsampling.

Simething tells me that this (5k 16x9 readout) will be one of the "super star" features of the new Phantom 5.

CT

 

[Dronefriend]

Yes, it is valid to call this a "4k" sensor. (Both, the Zoom and the Pro in crop HQ) However, I'm only saying that a 6k (100%+50%+50%) sensor readout can produce a nocicably higher frequency detail. Even using the same 4:2:0 chroma subsampling.

Simething tells me that this (5k 16x9 readout) will be one of the "super star" features of the new Phantom 5.

CT

Did the P4P not have this already?

 

[Roamer105]

Did the P4P not have this already?

Absolutely.

 

[Dronefriend]

Do you think the line skipping is hard coded in the Ambarella chip or in DJI firmware?

 

[Tiny]

I was using my MPP to film without flying, just holding the Mavic in my hand. I realized that the movement of the mavic didn’t exactly represent what I saw on the screen. I don’t just mean the gimbal movement, but also shaking the Mavic still produced smooth video. Is there some kind of digital picture stability? Wouldn’t it have to sample a larger FOV and reduce it to keep it stable. Is this why we could never get full sensor readout? Would the footage be shaky if it did?