Digital 2x and 4x extender

0lf wrote:

Ephemeris wrote:

0lf wrote:

Ephemeris wrote:

0lf wrote:

Ephemeris wrote:

drsnoopy wrote:

Ephemeris wrote:

drsnoopy wrote:

Ephemeris wrote:

drsnoopy wrote:

If something sounds too good to be true, it almost certainly is. A 4x “digital TC” is only using one sixteenth of the sensor area, so that’s a 1.5MP image. When that is upscaled to 24MP, you’re not going to have “image quality retained”. It’s a trick, pure and simple, more suitable for smartphones than cameras. But then I guess that’s the target demographic.

Maybe it's trying to say this is an improvement over a 4x crop. I think it's that which is the comparison rather than with 4x FL.

Same thing, same issue. 4x crop is the same as 4x “digital teleconverter”. A 24MP R50 sensor with a 4x crop = 1.5 MP image.

The question: is it the same thing. They suggest it isn't the same.

So in camera zoom isn't the same as the equivalent crop after the fact.

So no, not the same.

OK then please explain how it is “not the same”. The R50 has 24MP, no more. If you apply a 4x (which must be squared) you have just 1.5MP to use. If that is then upscaled, it is not using genuine data and there will be artefacts and loss of detail. You can’t change the laws of physics of maths.

I couldn't possibly comment on changing the laws of physics of maths but changing the laws (rules, guide?) Of maths and science is a daily affair.

So what I said is that they (Canon) suggest that taking a 4x crop after the fact isn't the same as what they (Canon) are doing with this digital zoom.

Noone has so far explained how this system operates.

If you take a 4x crop after the fact. The result is A

If I use Canons in camera zoom at a x4 level. The result is B

We display the two images at equal size

The suggestion (Canons) is that A does not equal B

I say again I do not know how this Canon system works, however I would be interested in knowing.

This was not meant to be a test of the feature, so there is variation between the shots, but may be you will find it informative

https://www.dpreview.com/forums/thread/4409591

Given the results with the G7x, I would presume the R50 to give excellent results at 2x.

Thanks Olf. I think your post is a little more helpful that the response you got in the past..

It does appear to be a helpful feature.

Thanks for your post I appreciate it

I just find an old test I made after buying the G7x2

max optical zoom + 2x crop in post

2x digital zoom

Thanks Olf.

Looks like the digital zoom has some noise, almost the type you get from creating a JPEG. What do you think?

yes, there is some artifact from the upscalling. But also it seems the sharpening is a bit more agressive with digital zoom (but imho, it is the cropped image that is lacking in this department). In definitive I think both are good enough to be usable.

I agree. I was looking at the cab of the crane.

Keep in mind this is from a 1inch 20Mp sensor, so not as clean as the R50 sensor

Thanks.

To add a bit mode data, below is the previously posted cropped photo upscalled by pixelmator photo

I think it's (the camera) doing a useful function.

2x cropped and pixelmator photo upscalling

Ephemeris wrote:

BirdShooter7 wrote:

chipman wrote:

Here's a sample of 2x digital zoom from 6 years ago -

800mm eq. from a FZ1000 - 1 inch 20 meg sensor.

Like I said above, about the same IQ as your basic cell phone.

Thanks for this but I thought we were discussing whatever version of the feature is in these latest R cameras and if it’s something more than a simple upscale.

We are. That is more or less the question.

I had thought whilst on my daily rounds if an OEM could use something at the level of Gigapixel AI becuase they now have a multi core AI engine within the ASIC it could be quite a selling point.

If course plenty will disagree with me (which is just fine) but more musing the possibility.

We are leveraging a new system from nVidia Drive Hyperon using a SoC called Atlan. Maybe Canon have something as advanced (this does use quite a bit of juice).

I’ve had similar thoughts and it seemed like it would be an interesting feature to me.

BirdShooter7 wrote:

Ephemeris wrote:

BirdShooter7 wrote:

chipman wrote:

Here's a sample of 2x digital zoom from 6 years ago -

800mm eq. from a FZ1000 - 1 inch 20 meg sensor.

Like I said above, about the same IQ as your basic cell phone.

Thanks for this but I thought we were discussing whatever version of the feature is in these latest R cameras and if it’s something more than a simple upscale.

We are. That is more or less the question.

I had thought whilst on my daily rounds if an OEM could use something at the level of Gigapixel AI becuase they now have a multi core AI engine within the ASIC it could be quite a selling point.

If course plenty will disagree with me (which is just fine) but more musing the possibility.

We are leveraging a new system from nVidia Drive Hyperon using a SoC called Atlan. Maybe Canon have something as advanced (this does use quite a bit of juice).

I’ve had similar thoughts and it seemed like it would be an interesting feature to me.

Indeed. If done well it's a powerful feature set. I'd also thought if it knows which lens and camera (so easy to have that data) it could upscale the sharper (above some limit that a user sets) portion from a lens give the particular F stop, and provide some Tele function.

-- hide signature --

Some of my bird photos can be viewed here: https://www.flickr.com/photos/gregsbirds/

RLight wrote:

I've used DC on the former PowerShots, its usefulness is limited. Pretty much if you're a JPEG shooter and don't want to /can't touch RAWs or crop in post, sure. And, I'll say the JPEGs on the R50 are pretty good. It's a smartphone-like option that is a gimmick to the point of others. But, I might argue it's actually useful in theory here. I'm pretty happy with the SOOC JPEGs from the R50, enough it could be "worth it". I may give DPP4 a shot now (on those R50 RAWs I have) that I figured out my lens profile issues... Not today though.

That's interesting. So it's got a set of functions that it's leveraging.

It's not "magic" though, it modifies things like your aperture a touch, really it's a no-post crop that does slightly alter your exposure/metering for the center of the image in my experience, but just crops it for you to the point of others, so it's more than what folks around here argue, less than what Canon insinuates, falls in the middle. That's not a bad thing if you intended to throw away the rest of the uncropped image anyways... Now sure, shoot RAW, do it yourself in post. But about that buffer... And folks coming from smartphones are used to this sort of thing too. Not inappropriate for the target audience of this thing. And again, those SOOC JPEGs are pretty good, it's not like you're gonna kick yourself for not being able to raise a shadow and then crop, or change WB and then crop on the R50. That's just not my experience in the past 48 hours with this thing. It's bang on with things like WB and exposure, right out of camera. So yeah, crop away in camera...

So it is doing more than you could with just a raw as it's adjusting F stop and metering if I understand correctly.

BTW, the real help the Digital TC does? Framing and metering. So as I said, it is somewhat useful.

I should add framing to the above features but sounds a positive option to have.

After all I use 1.6x crop on my R5 sometimes to help with AF, framing and just to help me see a little better.

RLight wrote:

I've used DC on the former PowerShots, its usefulness is limited. Pretty much if you're a JPEG shooter and don't want to /can't touch RAWs or crop in post, sure. And, I'll say the JPEGs on the R50 are pretty good. It's a smartphone-like option that is a gimmick to the point of others. But, I might argue it's actually useful in theory here. I'm pretty happy with the SOOC JPEGs from the R50, enough it could be "worth it". I may give DPP4 a shot now (on those R50 RAWs I have) that I figured out my lens profile issues... Not today though.

It's not "magic" though, it modifies things like your aperture a touch, really it's a no-post crop that does slightly alter your exposure/metering for the center of the image in my experience, but just crops it for you to the point of others, so it's more than what folks around here argue, less than what Canon insinuates, falls in the middle. That's not a bad thing if you intended to throw away the rest of the uncropped image anyways... Now sure, shoot RAW, do it yourself in post. But about that buffer... And folks coming from smartphones are used to this sort of thing too. Not inappropriate for the target audience of this thing. And again, those SOOC JPEGs are pretty good, it's not like you're gonna kick yourself for not being able to raise a shadow and then crop, or change WB and then crop on the R50. That's just not my experience in the past 48 hours with this thing. It's bang on with things like WB and exposure, right out of camera. So yeah, crop away in camera...

BTW, the real help the Digital TC does? Framing and metering. So as I said, it is somewhat useful.

Yes, having the extra magnification in the viewfinder can be useful, but the Digital TC concept is not necessary for that at all; the already-existing "crop mode" in Canon FF cameras does that, too.

The only thing that the upsampling really does, perhaps, is create a JPEG that handles subsequent resampling better than a simply-cropped JPEG or crop mode JPEG would.  You could do that yourself before distributing the files or uploaded them somewhere with unknown resampling quality, the camera just does it automatically with the DTC.  There may be a slight benefit to upsampling pixels before committing to JPEG, but the fact of the Bayer CFA does make that benefit quite small, compared to doing that with an image that was previously downsampled or came from a sensor without a CFA, like a 3-CCD video frame grab or a Foveon image.

I called it a gimmick because the name of the mode implies some kind of optical gain, but there is no optical gain.  Canon seems to want to sell these cameras to people based on the false premise that they have something optically better than crops going on, and as a Canon user I am very embarrassed by this deception.  "My 400mm f/8 lens becomes a real 1600mm and is still f/8!".  Many of us have been wishing for more options of crop mode size, or even a continuous crop mode with the zoom ring or control ring, but then they start giving this "raw-less, upsized-JPEG crop mode masquerading as a TC" nonsense instead.

John Sheehy wrote:

RLight wrote:

I've used DC on the former PowerShots, its usefulness is limited. Pretty much if you're a JPEG shooter and don't want to /can't touch RAWs or crop in post, sure. And, I'll say the JPEGs on the R50 are pretty good. It's a smartphone-like option that is a gimmick to the point of others. But, I might argue it's actually useful in theory here. I'm pretty happy with the SOOC JPEGs from the R50, enough it could be "worth it". I may give DPP4 a shot now (on those R50 RAWs I have) that I figured out my lens profile issues... Not today though.

It's not "magic" though, it modifies things like your aperture a touch, really it's a no-post crop that does slightly alter your exposure/metering for the center of the image in my experience, but just crops it for you to the point of others, so it's more than what folks around here argue, less than what Canon insinuates, falls in the middle. That's not a bad thing if you intended to throw away the rest of the uncropped image anyways... Now sure, shoot RAW, do it yourself in post. But about that buffer... And folks coming from smartphones are used to this sort of thing too. Not inappropriate for the target audience of this thing. And again, those SOOC JPEGs are pretty good, it's not like you're gonna kick yourself for not being able to raise a shadow and then crop, or change WB and then crop on the R50. That's just not my experience in the past 48 hours with this thing. It's bang on with things like WB and exposure, right out of camera. So yeah, crop away in camera...

BTW, the real help the Digital TC does? Framing and metering. So as I said, it is somewhat useful.

Yes, having the extra magnification in the viewfinder can be useful, but the Digital TC concept is not necessary for that at all; the already-existing "crop mode" in Canon FF cameras does that, too.

The only thing that the upsampling really does, perhaps, is create a JPEG that handles subsequent resampling better than a simply-cropped JPEG or crop mode JPEG would. You could do that yourself before distributing the files or uploaded them somewhere with unknown resampling quality, the camera just does it automatically with the DTC. There may be a slight benefit to upsampling pixels before committing to JPEG, but the fact of the Bayer CFA does make that benefit quite small, compared to doing that with an image that was previously downsampled or came from a sensor without a CFA, like a 3-CCD video frame grab or a Foveon image.

I called it a gimmick because the name of the mode implies some kind of optical gain, but there is no optical gain. Canon seems to want to sell these cameras to people based on the false premise that they have something optically better than crops going on, and as a Canon user I am very embarrassed by this deception. "My 400mm f/8 lens becomes a real 1600mm and is still f/8!". Many of us have been wishing for more options of crop mode size, or even a continuous crop mode with the zoom ring or control ring, but then they start giving this "raw-less, upsized-JPEG crop mode masquerading as a TC" nonsense instead.

It has been a few decades since I paid much attention to image scaling algorithms and I may not remember correctly or newer algorithms may be available. It makes sense to not do upscaling in camera when raw quality is selected since raw implies later processing on a bigger faster computer. If the goal is to get a JPEG from the camera quickly and avoid spending time post processing, then it seems to me that the upscaling in the camera is useful. But, if one is to do post processing anyway then one may choose algorithms that are slower or use more memory on a larger computer than the one in the camera.

As has been mentioned, it is always necessary to do some sharpening after scaling.

Comparison gallery of image scaling algorithms: https://en.wikipedia.org/wiki/Comparison_gallery_of_image_scaling_algorithms

The scaling algorithm that I usually use when down-scaling an image (default for GraphicsMagick) is: https://en.wikipedia.org/wiki/Lanczos_resampling

For upscaling the default algorithm for GraphicsMagick is: https://en.wikipedia.org/wiki/Mitchell%E2%80%93Netravali_filters

Gimp also allows one to choose the scaling algorithm which assumes one knows the tradeoffs among the various algorithms and can choose an algorithm based upon the desired result.

https://en.wikipedia.org/wiki/Image_scaling

http://www.graphicsmagick.org/GraphicsMagick.html
"
-filter <type>
use this type of filter when resizing an image
Use this option to affect the resizing operation of an image (see -geometry). Choose from these filters (ordered by approximate increasing CPU time):
Point Box Triangle Hermite Hanning Hamming Blackman Gaussian Quadratic Cubic Catrom Mitchell Lanczos Bessel Sinc
The default filter is automatically selected to provide the best quality while consuming a reasonable amount of time. The Mitchell filter is used if the image supports a palette, supports a matte channel, or is being enlarged, otherwise the Lanczos filter is used.
"

https://gmic.eu/reference/list_of_commands.html#geometry_manipulation

PicPocket wrote:

John Sheehy wrote:

PicPocket wrote:

BrianOdell wrote:

Canon states that this is not simply a crop, and retains quality (jpeg only).

I wonder what that means. Free pixels? Retains 50% quality? JPEG only, so some processing trick. I wonder if that can mean crop and upscale as best as we can

That statement is as subjective as it gets it, unless there is some more detail there

Some people have suggested some kind of AI upsizing, but at the speed that these images can be burst at, it doesn't seem likely that there can be any kind of real "intelligence".

When people start saying AI, it almost always starts to mean different things from context to context. The AI from 10 years ago is perhaps just automation/algorithm today. It's a fancy way to say we have a better algorithm. When it comes to learning models, with today's technology, it almost always means offloading to an online high capacity service, or spending ages running it local. There is no AI in a box as tiny as these cameras - that is, if we can even agree on what AI actually means

We use MAX78000 in very small packages (much smaller than an R) and they are optimized for deep convolutional neural networks and can do one and two dimensional convolution processing.

This is a true CNN so it would be said to be AI and AD advertise as such.

John Sheehy wrote:

RLight wrote:

I've used DC on the former PowerShots, its usefulness is limited. Pretty much if you're a JPEG shooter and don't want to /can't touch RAWs or crop in post, sure. And, I'll say the JPEGs on the R50 are pretty good. It's a smartphone-like option that is a gimmick to the point of others. But, I might argue it's actually useful in theory here. I'm pretty happy with the SOOC JPEGs from the R50, enough it could be "worth it". I may give DPP4 a shot now (on those R50 RAWs I have) that I figured out my lens profile issues... Not today though.

It's not "magic" though, it modifies things like your aperture a touch, really it's a no-post crop that does slightly alter your exposure/metering for the center of the image in my experience, but just crops it for you to the point of others, so it's more than what folks around here argue, less than what Canon insinuates, falls in the middle. That's not a bad thing if you intended to throw away the rest of the uncropped image anyways... Now sure, shoot RAW, do it yourself in post. But about that buffer... And folks coming from smartphones are used to this sort of thing too. Not inappropriate for the target audience of this thing. And again, those SOOC JPEGs are pretty good, it's not like you're gonna kick yourself for not being able to raise a shadow and then crop, or change WB and then crop on the R50. That's just not my experience in the past 48 hours with this thing. It's bang on with things like WB and exposure, right out of camera. So yeah, crop away in camera...

BTW, the real help the Digital TC does? Framing and metering. So as I said, it is somewhat useful.

Yes, having the extra magnification in the viewfinder can be useful, but the Digital TC concept is not necessary for that at all; the already-existing "crop mode" in Canon FF cameras does that, too.

The only thing that the upsampling really does, perhaps, is create a JPEG that handles subsequent resampling better than a simply-cropped JPEG or crop mode JPEG would. You could do that yourself before distributing the files or uploaded them somewhere with unknown resampling quality, the camera just does it automatically with the DTC. There may be a slight benefit to upsampling pixels before committing to JPEG, but the fact of the Bayer CFA does make that benefit quite small, compared to doing that with an image that was previously downsampled or came from a sensor without a CFA, like a 3-CCD video frame grab or a Foveon image.

I called it a gimmick because the name of the mode implies some kind of optical gain, but there is no optical gain. Canon seems to want to sell these cameras to people based on the false premise that they have something optically better than crops going on, and as a Canon user I am very embarrassed by this deception. "My 400mm f/8 lens becomes a real 1600mm and is still f/8!".

No John. It's specifically called DIGITAL and not Optical.

I think the problem isn't with Canon here at all and it's not deception but possibly biased confusion.

Let's not tar a manufacturer incorrectly that's not positive.

Many of us have been wishing for more options of crop mode size, or even a continuous crop mode with the zoom ring or control ring, but then they start giving this "raw-less, upsized-JPEG crop mode masquerading as a TC" nonsense instead.

It's not nonsense though, far from it. If it's not for you that's just one of those things. I however am interested in finding out more about how this feature works.

-- hide signature --

Beware of correct answers to wrong questions.
John
http://www.pbase.com/image/55384958.jpg

RLight wrote:

The center of the lens is always sharper, so long as your lens isn’t defective or damaged, yes.

The center of the image circle is usually the sharpest, but the corners of a crop can be softer than the corners of the entire sensor, when both are normalized to the same displayed size.  It is a lens-by-lens thing.  The center of the image will never be sharper in a normalized crop, though; it will always be softer.  That is how I think of it in terms of normalizing for the same image.

Of course, if you are using a mode that uses a fraction of the sensor just to avoid wasted MBs, and would crop from the full sensor, otherwise, then this whole normalization perspective has no practical value.

People need to be aware that crop modes, especially when used with zooms, and especially with fixed-open-f-ratio zooms, can be extremely counterproductive.  Let's say you have a 24-105 fixed-f/4 zoom.  If you use 25mm in 4x TC mode vs 100mm in full-sensor mode, both at f/4 on a 24MP sensor, the 100mm gives you 100/4 with 24MP, but the 25mm and 4x "TC" gives you 1.5MP at a FF equivalent of 100/16.

It is good to be aware and not use a crop mode or fake firmware TC when you don't really need them.  Even a real 4x TC would yield 100/16; the only benefit would be 24MP used.

davesurrey wrote:

A general comment:

I’ve just seen this post and I have to say I disagree with those who are quite adamant that a 4x crop must give 1.5MP resolution. This doesn’t seem to be thinking out the box or considering what DSP technology can do these days.

Here’s a simple example. When CDs first came out I happened to work for Sony and so could get one of their CD players very cheap with staff-sales but I bought one from Philips instead.

Friends were baffled as the Sony used a 16 bit DAC whereas the Philips “only” used a 14 bit DAC. They just couldn’t believe it would sound as good, let alone better, until they heard it for themselves.

The reason was that the Philips used 4x over-sampling so could use a more linear 14 bit DAC (16 bits DACs were not so linear in those days) and the anti-aliasing or reconstruction filters could have a much more gradual pass band leading to far less phase distortion/ripples in the pass band. As a result they just sounded better.

It's rarely wise to underestimate what engineers can do, which may appear to “cheat” the laws of physics.

That doesn't seem like an applicable analogy.  The sensor would have to be read multiple times in a row, either slowing the rolling shutter time or doing low-bitdepth readout, to curb noise a bit (eliminate far outliers), but that will not increase resolution.  You'd have to shift the sensor to multiple positions, too, to increase capture resolution beyond 1.5 million samples to 24 million, and shifting to 16 positions would be very time consuming.