JPEG or RAW - a new look.

Have something like Adobe Bridge and you can see RAW files prior to processing and Jpegs and Tiffs. Just because Jpegs are what is used as the final result most of the time doen't mean that that should be what you start with. It is a choice obviously.

Any software that can make a lot of changes to an image without user intervention doesn't seem like a good idea to me. No matter how 'intelligent' software is it can make mistakes. What if an image doesn't need sharpening at all! Again it is a choice obviously, but I like to edit/process each image with regards to what I think needs changing, if indeed anything needs changing. ;)

And there are those that don't seem to want to do what most people do that gives great results, and seem to go out of their way to do something different which may or may not be an improvement for whatever reason. And then seem to want to convert others to their way of thinking. :rolleyes:

It is not denying that technology can do some amazing things, as photographers we use technology all the time, but sometimes things are embraced either before they are ready, or they are 'fixing' a problem that didn't need 'fixing' for most people.

All I will say is the simple fact that if you haven't tried it you can't judge it.
 
All I will say is the simple fact that if you haven't tried it you can't judge it.
My post was more a comment about you, and the need to seemingly go out of your way to do something different to everyone else, and the evangelise whatever you find. I have no need to recover data, or should I say, in some cases, make up data, from Jpeg as I don't take images in the Jpeg format, and haven't for over a decade. For me the RAW file is the option to have the most data recorded, so why wouldn't I use it. :thinking:

When memory was expensive (for cameras and computers), computers had less power, and cameras were slower (for fps and shallower buffer by using the RAW format), I could understand the use of choosing the Jpeg format. It was mentioned in the other thread that 'many pro photographers' shoot in Jpeg as some kind of justification for using the Jpeg format, but in most instances that would be sport photographers who were under time and technology restrictions to deliver images quickly. I doubt many non sport pro photographers (if you want to use them as a yardstick) would choose to shoot in the Jpeg format if they are under no time pressure to deliver images. There will always be a few of course who stick to Jpeg no matter what. Good luck with that. (y) :)
 
The software may be quite good but garbage in garbage out comes to mind - Jpeg is fine for many things especially if you don't intend to edit, and it's also perfect as a final edited file, but no matter which way you argue it doesn't hold the full amount of data that a raw file does and you aren't going to be able to pull out the same dynamic range. Shoot raw for the full benefits of raw.
 
Hi - new member here. I'm an Oly m43 user. I saw a similar post to this over on the UK e-system user group. The software seems a bit pointless to me. Once you've done all the things that a camera does in making a jpeg (sharpening, NR, tone curve, WB, compression etc), then you can never get the original data back.

I accept that if these aspects have been lightly done in camera then you can probably get close, but then you could do whatever changes/edits are needed directly on the jpeg using LR or similar. So, what's the point?
 
Last edited:
........but then you could do whatever changes/edits are needed directly on the jpeg using LR or similar. So, what's the point?
Welcome Paul.....
The point? Unless you're going to get into some serious masking then LR or similar tends to apply global changes to the image whereas this Topaz utility would appear to analyse small portions of the image and apply specific changes before creating the quasi-raw file.

Bob
 
Hi - new member here. I'm an Oly m43 user. I saw a similar post to this over on the UK e-system user group. The software seems a bit pointless to me. Once you've done all the things that a camera does in making a jpeg (sharpening, NR, tone curve, WB, compression etc), then you can never get the original data back.

I accept that if these aspects have been lightly done in camera then you can probably get close, but then you could do whatever changes/edits are needed directly on the jpeg using LR or similar. So, what's the point?

Well apart from recovering lost detail in a JPEG it also reduces noise almost perfectly, as shown on my Flick site.

And you can batch process hundreds of photos all at once.

A few years ago I took about 3,000 photos at Millstreet Horse Show in Ireland but accidentally under exposed most of them, making it really difficult to process them

This software has almost totally removed a ton of noise from them without removing detail and also recovered a load of them where I just missed focus.

So far I've put through about 2,500 of them so it does a huge amount for the cost of it.

As usual I would say try it out.
 
Well apart from recovering lost detail in a JPEG

I'm still perplexed over this one statement.

It's lost in a JPG - it's lost, nothing there, gone, never written, binary 0000 0000, 0x00 ..... end of!

So, how can it be recovered? It can be invented or created but, NEVER recovered!
 
You can start with a Jpeg or a raw file But...

any "Amazing Software" will always be able to do more with a raw file.
 
When it comes to AI things are now totally different.
Since every pixel is influenced by the nearby pixels the AI can infer what was there even when there is nothing there - think of Conway's Game of Life.
And if detail is missing it can be recreated since the AI has learnt on millions of images.
So the pixels are subjected to trillions of calculations to produce the recovered image.
But it does make mistakes.
Because it is, if you like, still very young but growing better with every day.
With the advent of AI photography has now changed irrevocably.
 
And if detail is missing it can be recreated since the AI has learnt on millions of images.
So the pixels are subjected to trillions of calculations to produce the recovered image

I'm glad you like it so much - happy for you.

But I'm struggling. You're using terms like 'recovered' and 'recreated'.

It was never there in the original - it's neither recreated nor recovered - it simply was never there in the first place!

Do you know what (and I'll concede one point)? I think it could have been there 'originally' but the JPG algorithm threw it away before it got written to storage!

Is it getting to the point that I might just as well buy any ol camera capable of writing JPGs and send it out on its own adventure?
 
I'm glad you like it so much - happy for you.

But I'm struggling. You're using terms like 'recovered' and 'recreated'.

It was never there in the original - it's neither recreated nor recovered - it simply was never there in the first place!

Do you know what (and I'll concede one point)? I think it could have been there 'originally' but the JPG algorithm threw it away before it got written to storage!

Is it getting to the point that I might just as well buy any ol camera capable of writing JPGs and send it out on its own adventure?

Well autonomous cars are now on the roads in America:

https://www.wired.com/story/waymo-self-driving-this-week-cars/

And with these new programs my old 350D and 40D etc have a totally new lease of life.

But you're still thinking in human terms.

Hal (which is what I mentally call the AI's, for obvious reasons) doesn't - all he knows is data, numbers, huge datasets.

And he is not bound by our terminology, to him a JPEG or a TIFF or a RAW file is all the same - numbers.

Which he can manipulate at incredible speeds, to produce what to us is an image.
 
When it comes to AI things are now totally different.
...
With the advent of AI photography has now changed irrevocably.

AI hasn't just happened, and AI isn't really a 'thing', it's just a collection of approaches that, thanks to ever increasing processing speed, and availability of training material, is more readily commoditised. I was working with neural nets 30+ years ago.

There's proactive 'AI' in the sense of Sony's eye AF (other manufacturers are available!), which uses a set of specific techniques to identify an eye in an image. This has certainly made the process of taking a well focussed image easier, but I don't think it's changed the core tenets of photography.

Matrix / Evaluative metering is a form of 'AI' in this sense too and that's been around for donkey's years. Just makes things easier, but doesn't change anything fundamental.

There's computational photography, another set of core algorithms, processes - from the basic Panorama functions available in many cameras, to the auto scene selections, and portrait backdrop blurs etc too, computational ND filter etc. Some years old, some new. They are tools you use when you need to to get a result you want. They are shortcuts that emulate (usually worse) something you can achieve better through normal routes. They are convenience functions that give you something near to what you want for little or no effort. There is scope for this areas to be huge though I think, especially as you start to combine multiple lenses / sensors into a single image), that's when it may surpass photography as we know it today.

There's retrospective 'AI', as in this case, which uses another set of tools and techniques to analyse images. There's some horrible crap out there (plastic faces etc) and there's some useful stuff there too. Again in most cases they are simply making a stab at automating something that you can do better if you spend the time yourself. The results won't be better - AI isn't going to more accurately analyse an image than you are, it just does it with less effort and that's a compromise you accept.

You'd still strive for the best you can, and use the best tools available when you get things wrong (or conditions don't allow you to do any better). This latest set of tools may help in those circumstances, but I really can't see them being in any standard workflow.
 
Well autonomous cars are now on the roads in America:

https://www.wired.com/story/waymo-self-driving-this-week-cars/

And with these new programs my old 350D and 40D etc have a totally new lease of life.

But you're still thinking in human terms.

Hal (which is what I mentally call the AI's, for obvious reasons) doesn't - all he knows is data, numbers, huge datasets.

And he is not bound by our terminology, to him a JPEG or a TIFF or a RAW file is all the same - numbers.

Which he can manipulate at incredible speeds, to produce what to us is an image.
The quality of the data set still matters.
Higher quality data > Lower quality data when it comes to analysing an image.

Please don't fall into the marketing trap that says AI can make something fantastic out of rubbish. It will work a lot better if you feed it better quality material than if you feed it junk. It does not make a lower bit resolution image equal to or better than a higher one (within the limits of dynamic range etc).
 
This latest set of tools may help in those circumstances, but I really can't see them being in any standard workflow.

Really? Well I have used Neat Image in my workflow for many years because it was a good noise reduction system, which any images which need NR would be put though first.

And they could be batch processed.

But it needed special "profiles" needed of each camera at every ISO the camera could use.

This was then utilised by the program to reduce noise in a specific camera at a specific ISO.

This was done automatically in the program so if you had a number of photos taken at different ISOs with different cameras this was no problem and it could produce decent results.

JPEG to RAW beats it hollow in NR alone - just put an image in, choose normal or high for the Noise and Blur setting, and that's it!

Just one of the reasons I paid for it.
 
Last edited:
The quality of the data set still matters.
Higher quality data > Lower quality data when it comes to analysing an image.

Please don't fall into the marketing trap that says AI can make something fantastic out of rubbish. It will work a lot better if you feed it better quality material than if you feed it junk. It does not make a lower bit resolution image equal to or better than a higher one (within the limits of dynamic range etc).

Well Hal has been fed on millions of images and has learned.

But if you want to believe in what has been true for years, please do so.

But just remember - "do what you've always done - get what you've always got"

And unlike quite a few who criticise this program without even trying it, I am actually using it, and it has now processed literally thousands of images already so I do know what I'm talking about!
 
Well Hal has been fed on millions of images and has learned.

But if you want to believe in what has been true for years, please do so.

But just remember - "do what you've always done - get what you've always got"

And unlike quite a few who criticise this program without even trying it, I am actually using it, and it has now processed literally thousands of images already so I do know what I'm talking about!

Sorry but this is just wrong. In this context 'AI' will give you the best it can for a given input. Give it a better input and it will give you a better output.

To think that 'AI' has magically somehow managed to surpass that basic truth demonstrates a lack of understanding.

I think the progress is great, I think it has many uses, and I think it will continue to improve, but it won't change the fact that starting with better data will yield a better result. This isn't the raw v jpeg argument - I think in many many cases a jpeg contains more than sufficient information to yield great images even when pushed, this is simply to make the point that when you are talking about using a tool to recover an image that is already struggling (high noise etc), more 'original' information is better than less.

By the way, there is nothing magical about AI. It's well understood in its many forms.
 
Last edited:
Sorry but this is just wrong. In this context 'AI' will give you the best it can for a given input. Give it a better input and it will give you a better output.

To think that 'AI' has magically somehow managed to surpass that basic truth demonstrates a lack of understanding.

I think the progress is great, I think it has many uses, and I think it will continue to improve, but it won't change the fact that starting with better data will yield a better result. This isn't the raw v jpeg argument - I think in many many cases a jpeg contains more than sufficient information to yield great images even when pushed, this is simply to make the point that when you are talking about using a tool to recover an image that is already struggling (high noise etc), more 'original' information is better than less.

By the way, there is nothing magical about AI. It's well understood in its many forms.

No doubt at all in what you say about better input

BUT:

https://petapixel.com/2017/10/31/website-uses-ai-enhance-photo-csi-style/

https://www.wired.co.uk/article/google-brain-ai-pictures-blur

https://www.fastcompany.com/9014977...le-pixelated-photos-into-crystal-clear-images

This program is simply a better version of some of the above

But this one CAN create something from nothing:

https://www.theverge.com/2018/12/17...e-nvidia-generative-adversarial-networks-gans
 
That last one is an example of a specialist application - it's truly impressive, but it's not part of Topaz, and I'd say that's digital art rather than photography. For years it's been possible to generate natural looking foliage trough the mathematics of fractals so again it's not new, but it is amazing.

It does kind of illustrate the point as to why, if it were available at a consumer level, you would not want to routinely apply that to all your images - hypothetically of course! When Aunty Mable stops looking like Aunty Mable, is it still a photograph? That said, those of a Facebook / Snapchat persuasion have already gone well beyond that :)
 
I am going through the process of putting thousands of images through this software and have discovered something that seems to support Topaz' claim:

"JPEG to RAW’s machine learning models expand the sRGB colorspace to ProPhoto RGB, which is even better than a regular RAW file! This works so well because there is usually enough information in at least one color channel for our models to reconstruct missing detail with a high level of preciseness. "

And it seems to be true!

I have a program called Fast Picture Viewer which allows me to cull photos very rapidly - and it puts a lot of info on the screen - one of them being the colour space the photo is in.
Normally with a JPEG or TIFF from a JPEG this is sRGB but with the TIFFs from J2R it identifies the colour space as ProPhoto RGB, exactly what Topaz claims!
 
Normally with a JPEG or TIFF from a JPEG this is sRGB but with the TIFFs from J2R it identifies the colour space as ProPhoto RGB, exactly what Topaz claims!
Sorry, but that proves nothing... You can convert a TIFF to whatever colour space you want it does not mean that it is utilising the available gamut of that colour space...
 
This works so well because there is usually enough information in at least one color channel for our models to reconstruct missing detail with a high level of preciseness.

Oh, the advertising claims of this - it's driving me nuts!

One channel of three - the result of mixing red/green or red/blue gives a big difference yet they claim they can do it from just one channel - so, are they guessing the missing bits?

They can expand (unknown/lost) data into a different colourspace. Hardly surprising; you can always put more water into a halfempty bucket no matter how you make it up or invent.

"You can reconstruct missing detail" - sorry, how do you know what to construct if it's missing? What do you mean it was a giraffe I photographed, I thought it was an elephant so that's what I reconstructed (I know I've taken this to an obtuse level but I think you get my drift)

And the word is PRECISION :p

Actually, if this was printed matter in Photojournal in this country it comes close to contravening Advertising Standards legislation
 
And it seems to be true!

It isn't. It can't change the colourspace of a file. I can stick a Ferrari badge on the front and back of my car but that doesn't make it a Ferrari.

Same here, just because it says the file is in a different and better colourspace doesn't mean it actually is.

Actually, if this was printed matter in Photojournal in this country it comes close to contravening Advertising Standards legislation

This. There are so many ridiculous claims about this piece of software, it just sounds like a scam to get people to part with $100 or however much it is.
 
Well regardless of all the negativity about it, and especially the use of terms like RAW and PhotoRGB the simple fact is they have NOT been prosecuted or, as far as I know, even reported to the relevant authorities, either here or in America, for false advertising.

And in America the FTC really loves to prosecute people so I will go on accepting what they say as the truth, unless, and until, someone can prove them wrong, since I have already proven, and shown, that a large amount of their so called "scam" claims are actually true!

But you are very welcome to believe exactly what you wish, just as I am.
 
since I have already proven, and shown, that a large amount of their so called "scam" claims are actually true!
I have a program called Fast Picture Viewer which allows me to cull photos very rapidly - and it puts a lot of info on the screen - one of them being the colour space the photo is in.
Normally with a JPEG or TIFF from a JPEG this is sRGB but with the TIFFs from J2R it identifies the colour space as ProPhoto RGB, exactly what Topaz claims!
You have proven nothing with regards colour space... that claim is simply a load of rubbish, I can convert a JPEG file to a TIFF then convert the colour space to ProPhotoRGB without using this piece of software and your Fast Picture Viewer would show exactly the same, just because the space has been converted to ProPhoto does not mean that the file is using the gamut of that new space...
 
I did this in 30 seconds exporting from Lightroom. A .jpg file with a ProPhoto colour space. Nothing new here and I saved £100+, also, best of all, I shot this in raw so actually have ALL the raw data with no additional, over priced, over hyped software at all.

Screenshot 2019-02-12 at 15.20.24.jpg20190206112621-Edit.jpg
 
I think Topaz Labs themselves are rather more realistic about the whole thing than some of the more excited comments posted here (Topaz: "JPEG to RAW AI helps you recover a portion of this lost dynamic range by bringing back lost shadow and highlight details. While you won't be able to recover completely missing details, it can work with even very small amounts of information."). Their example conversion (middle) shows obvious loss of edge detail compared to true raw (right):

https://topazlabs.com/static/jpeg-to-raw-ai/img/compression-bna.png

Those fibres have been interpreted as noise and removed. That's still an impressive recovery of a jpeg with such heavy artefacts, and the output would presumably look much better with a high quality, lightly compressed jpeg. But in that case, of course, a tool like this may well be superfluous. If you have a lot of images from a phone or an older camera with obvious artefacts, it may be a worthwhile investment, both for doing a decent job of pre-processing the images and for the convenience generating a 'raw' file that can be used alongside true raws in the same workflow. But I suspect for many of us this is a solution in search of a problem.
 
Does this conversion process ruin the flexibility that an original raw would provide
 
Does this conversion process ruin the flexibility that an original raw would provide

Well, it started off as a JPG so never had any comparable flexibility in the first place - think compromise from start to finish and I think you have your answer ;)
 
Back
Top