JPEG or RAW?

[Slobodan Blagojevic]

Are you volunteering to do this for us?

Ha! I already ruined one Netflix evening trying to download the damn program. It is like half a gygabite and slowed my Netflix to a crawl. Worse, it froze it every 5 minutes 😉

[Mark D Segal]

Ha! I already ruined one Netflix evening trying to download the damn program. It is like half a gygabite and slowed my Netflix to a crawl. Worse, it froze it every 5 minutes 😉

AH, good (I mean not the trouble, but the fact you are downloading it). Once this painful download completes and assuming it installs, should we look forward to your take on the results?

[Slobodan Blagojevic]

... should we look forward to your take on the results?

Maybe, but I can't promise. I am usually not the guy to shoot brick walls and test targets for fun.

[fdisilvestro]

The real debate is whether whatever "A.I." Topaz uses in its application is good enough to ingest a JPEG file and output a result that has equivalent photographic quality to a well-processed true raw file.

My answer to that is NO, the output of Topaz Jpeg to Raw is not at the level of a true raw file, understanding "true raw" as the original raw from which the JPEG was produced.

I would change the debate to:

whether whatever "A.I." Topaz uses in its application is good enough to ingest a JPEG file and output a result that has better quality than any other current way to edit a JPEG file. (and I don't have the answer to this)

Is the output a raw file? First consider that what this application claims to do is not a simple file conversion, but apply A.I. and try to recreate missing information that is coherent to the JPEG file, not necessarily to the real scene that was photographed. As a reference, consider the Google photos A.I. ability to colorize B&W images: would the colors be accurate to the original scene? Not necessarily, but the color information was "recreated" and is believable.

Having said that, the output file has some characteristics of a raw file: Linear raw, 16 bit per channel, uncompressed data, linear tone curve; while other characteristics are as in rendered files: color space encoded (ProPhotoRGB).

If you consider the processing of an image from "true raw" to rendered in tiff as a series of steps, I would say that the output from Topaz JPEG to RAW is just one step short of TIFF, so even if you can open the DNG in any raw processing application, there are some limitations to what you can do. For instance, in Lightroom/ACR there is no difference processing these DNGs instead of TIFFs. You cannot apply a DCP profile nor adjust white balance as in a true raw.

One more observation: JPEG to RAW tries to maximise detail by sharpening aggressively, so if you open the DNG with Capture one with its default settings the image will look really oversharpened. This is another big difference to "true raws", which in my experience are softer than processed images, especially if you use a camera with low pass filter.

Following the results of a test of recovering shadows (you may call it contrived)
Photo taken with a Nikon D800 in Raw + Jpeg mode, 14 bit raw and large, fine, optimal quality JPEG. Underexposed 6 stops.

The first image shows how the JPEG out of camera looks:



The following is what I could do with the "True raw" (NEF) file in LR:



And what I was able to do with the DNG output from JPEG to RAW + LR

[nirpat89]

My answer to that is NO, the output of Topaz Jpeg to Raw is not at the level of a true raw file, understanding "true raw" as the original raw from which the JPEG was produced.

I would change the debate to:

whether whatever "A.I." Topaz uses in its application is good enough to ingest a JPEG file and output a result that has better quality than any other current way to edit a JPEG file. (and I don't have the answer to this)

Is the output a raw file? First consider that what this application claims to do is not a simple file conversion, but apply A.I. and try to recreate missing information that is coherent to the JPEG file, not necessarily to the real scene that was photographed. As a reference, consider the Google photos A.I. ability to colorize B&W images: would the colors be accurate to the original scene? Not necessarily, but the color information was "recreated" and is believable.

Having said that, the output file has some characteristics of a raw file: Linear raw, 16 bit per channel, uncompressed data, linear tone curve; while other characteristics are as in rendered files: color space encoded (ProPhotoRGB).

If you consider the processing of an image from "true raw" to rendered in tiff as a series of steps, I would say that the output from Topaz JPEG to RAW is just one step short of TIFF, so even if you can open the DNG in any raw processing application, there are some limitations to what you can do. For instance, in Lightroom/ACR there is no difference processing these DNGs instead of TIFFs. You cannot apply a DCP profile nor adjust white balance as in a true raw.

One more observation: JPEG to RAW tries to maximise detail by sharpening aggressively, so if you open the DNG with Capture one with its default settings the image will look really oversharpened. This is another big difference to "true raws", which in my experience are softer than processed images, especially if you use a camera with low pass filter.

Following the results of a test of recovering shadows (you may call it contrived)
Photo taken with a Nikon D800 in Raw + Jpeg mode, 14 bit raw and large, fine, optimal quality JPEG. Underexposed 6 stops.

What do you get if you took the jpg out of the camera (image #1,) convert to Prophoto 16 bit tiff, put it in LR and/or PS, and give it the best treatment possible.  How would it compare with the JPG to RAW + LR output (image #3)?   How big of a difference?

[Mark D Segal]

My answer to that is NO, the output of Topaz Jpeg to Raw is not at the level of a true raw file, understanding "true raw" as the original raw from which the JPEG was produced.

I would change the debate to:

whether whatever "A.I." Topaz uses in its application is good enough to ingest a JPEG file and output a result that has better quality than any other current way to edit a JPEG file. (and I don't have the answer to this)

..............

Given the context set by the Topaz claim, I wouldn't change my question; rather I would call your's another equally valid question.

The three examples you show are very informative in respect of my question, clearly showing that there is acceptable photographic adjustment potential, at least in difficult raw files, that far exceed what one can do with a JPEG in the Topaz application. Again this is a sample of one, but nonetheless of indicative value.

[Bart_van_der_Wolf]

What do you get if you took the jpg out of the camera (image #1,) convert to Prophoto 16 bit tiff, put it in LR and/or PS, and give it the best treatment possible.  How would it compare with the JPG to RAW + LR output (image #3)?   How big of a difference?

Exactly, that's the test to perform. Does this application allow to produce better results from JPEG input than a straight JPEG edit? Is the J2R preprocessed JPEG more robust for postprocessing than the original JPEG?

I think we can only judge that for general subjects when a better A.I. model is trained, with less noise suppression and less sharpening applied than the 'Normal' setting does. I'll repeat, the same was the case with Gigapixel A.I., the initial models were too aggressive in noise suppression, but the current 'None' setting does very well for reasonable quality input.

Cheers,
Bart

[TomFrerichs]

…
The real debate is whether whatever "A.I." Topaz uses in its application is good enough to ingest a JPEG file and output a result that has equivalent photographic quality to a well-processed true raw file.
 …

Mark, I am aware that several posters have suggested that you have correctly summarized the question. I don't think you have, though.

The (admittedly overblown) claims made by Topaz marketing say: Use machine learning to convert JPEG to high-quality RAW for better editing. Prevent banding, remove compression artifacts, recover detail, and enhance dynamic range. The application is supposed to process a JPEG file into a form that is more amenable to further processing by reducing artifacts, recover highlight and shadow detail, etc. Nowhere is there a claim that it will result in a "like raw" image, only that the output would have fewer artifacts, better detail in highlight and shadow areas, and be less susceptible to  additional editing problems like posterization.

I think the proper question to ask, based upon what Topaz claims, is this: Does a JPEG file processed in this application exhibit fewer compression artifacts, have recovered or enhanced detail, and is more tolerant of additional editing compared to the original.

TG Frerichs

[faberryman]

The (admittedly overblown) claims made by Topaz marketing say: Use machine learning to convert JPEG to high-quality RAW for better editing. Prevent banding, remove compression artifacts, recover detail, and enhance dynamic range. The application is supposed to process a JPEG file into a form that is more amenable to further processing by reducing artifacts, recover highlight and shadow detail, etc. Nowhere is there a claim that it will result in a "like raw" image, only that the output would have fewer artifacts, better detail in highlight and shadow areas, and be less susceptible to  additional editing problems like posterization.

What part of their saying "Use machine learning to convert JPEG to high quality RAW" don't you understand. It is that false claim that has people's hackles up. You gloss over it.

[digitaldog]

What part of their saying "Use machine learning to convert JPEG to high quality RAW" don't you understand. It is that false claim that has people's hackles up. You gloss over it.

What's more, they are VERY clear about what they claim:


Edit JPEG as if you shot it in RAW

As IF you shot it in raw!
And yet, those of us who shoot raw can easily see that if (as an example), one's WB is utterly wrong, it has zero effect on processing that raw, yet where and when will we see an incorrectly captured JPEG with respect to WB be fixed as if it were shot in raw? ?


The proof of the lie (again):

[TomFrerichs]

What part of their saying "Use machine learning to convert JPEG to high quality RAW" don't you understand. It is that false claim that has people's hackles up. You gloss over it.

If you have been reading the entire thread, you would have noticed that in reply #58, Mr. Segal began with: I think it's time to call a truce - or even an armistice - on whether that application produces raw files. It doesn't, and Topaz is remiss in its representation of the product. I didn't include that in my quotation of his post in the interest of saving space.

I don't think anyone here will disagree that there's no way any program can create an original camera sensor capture (if that's the meaning of a raw) from a JPEG. Topaz marketing is nonsense.

Yes, I glossed over that. At this point I think we want an answer to Mr. Segal's question posed in #58 or my restatement. In other words, it's not a question of me ignoring the marketing idiocy but of getting past it to ask other questions.

As I was typing this, I noticed that Andrew was replying.  In response...I don't care how they do it, with artificial intelligence or fairy dust, or how they market it. They can claim that they use alien technology found on a crashed spaceship in Roswell, New Mexico. I don't give a damn if their marketing is dodgy. 

All I really want to know is this: Does the program output a file that can be edited better than the original JPEG? Can you answer that question by direct experience?

TG Frerichs

[digitaldog]

I don't think anyone here will disagree that there's no way any program can create an original camera sensor capture (if that's the meaning of a raw) from a JPEG. Topaz marketing is nonsense

Not so, read the OP's confused idea about DNG. Try reading a lack of admission from some, after multiple questions with yes or no answers IF the product converts JPEG to raw.

[albytastic]

If you have been reading the entire thread, you would have noticed that in reply #58, Mr. Segal began with: I think it's time to call a truce - or even an armistice - on whether that application produces raw files. It doesn't, and Topaz is remiss in its representation of the product. I didn't include that in my quotation of his post in the interest of saving space.

I don't think anyone here will disagree that there's no way any program can create an original camera sensor capture (if that's the meaning of a raw) from a JPEG. Topaz marketing is nonsense.

Yes, I glossed over that. At this point I think we want an answer to Mr. Segal's question posed in #58 or my restatement. In other words, it's not a question of me ignoring the marketing idiocy but of getting past it to ask other questions.

As I was typing this, I noticed that Andrew was replying.  In response...I don't care how they do it, with artificial intelligence or fairy dust, or how they market it. They can claim that they use alien technology found on a crashed spaceship in Roswell, New Mexico. I don't give a damn if their marketing is dodgy. 

All I really want to know is this: Does the program output a file that can be edited better than the original JPEG? Can you answer that question by direct experience?

TG Frerichs

I already have - on my Flickr site - and at the present moment am processing several hundred JPEGs using this program and every one is producing better results than I had before.

[Ray Harrison]

I already have - on my Flickr site - and at the present moment am processing several hundred JPEGs using this program and every one is producing better results than I had before.

Can you be specific on what you mean by "better than before"? Are you processing any of them afterwards?

[Alan Goldhammer]

I already have - on my Flickr site - and at the present moment am processing several hundred JPEGs using this program and every one is producing better results than I had before.

While this might be a true statement for JPEGs it is a totally false statement when one is talking about RAW files.  An 8 bit file sacrifices a considerable amount of data relative to a 14 bit file (assuming correct WB of the two).  The fair trial is to capture a RAW and high def JPEG on the same exposure.  Do the AI transformation on the JPEG and then compare how the two images can be processed.  My Nikon Z 6 can do this type of capture though I have no intent on doing any comparative trials as I can't be bothered to download another piece of software whose marketing claim bypasses the normal laws of physics.  I know that Thom Hogan has looked at both RAW and JPEGs in his new manual on the new Nikon Z cameras and has shown that there is a difference in quality, particularly around edges.

[albytastic]

Can you be specific on what you mean by "better than before"? Are you processing any of them afterwards?

Of course I am, and a few previous ones are on my Flickr site:

https://www.flickr.com/photos/20926615@N05/albums/72157678279709128

[albytastic]

While this might be a true statement for JPEGs it is a totally false statement when one is talking about RAW files.  An 8 bit file sacrifices a considerable amount of data relative to a 14 bit file (assuming correct WB of the two).  The fair trial is to capture a RAW and high def JPEG on the same exposure.  Do the AI transformation on the JPEG and then compare how the two images can be processed.  My Nikon Z 6 can do this type of capture though I have no intent on doing any comparative trials as I can't be bothered to download another piece of software whose marketing claim bypasses the normal laws of physics.  I know that Thom Hogan has looked at both RAW and JPEGs in his new manual on the new Nikon Z cameras and has shown that there is a difference in quality, particularly around edges.

So, in other words, and like a lot of people, you condemn this program without even trying it out?

Seems a bit ridiculous to me.

[digitaldog]

So, in other words, and like a lot of people, you condemn this program without even trying it out?

Seems a bit ridiculous to me.

He condemns nothing and simply provides facts you refuse to accept. Like your inability to accept DNG may not equal raw! That seems ridiculous to many Here.

[Mark D Segal]

Mark, I am aware that several posters have suggested that you have correctly summarized the question. I don't think you have, though.

The (admittedly overblown) claims made by Topaz marketing say: Use machine learning to convert JPEG to high-quality RAW for better editing. Prevent banding, remove compression artifacts, recover detail, and enhance dynamic range. The application is supposed to process a JPEG file into a form that is more amenable to further processing by reducing artifacts, recover highlight and shadow detail, etc. Nowhere is there a claim that it will result in a "like raw" image, only that the output would have fewer artifacts, better detail in highlight and shadow areas, and be less susceptible to  additional editing problems like posterization.

I think the proper question to ask, based upon what Topaz claims, is this: Does a JPEG file processed in this application exhibit fewer compression artifacts, have recovered or enhanced detail, and is more tolerant of additional editing compared to the original.

TG Frerichs

The Topaz problem starts with the first phrase you recited: "Use machine learning to convert JPEG to high-quality raw for.....". Machine learning cannot do this, so it is a false claim. Their claim is also that their application will give similar results to a well-processed raw file from a JPEG file. That was the basis of my definition of the issue, and I remain comfortable thinking it's valid to put it that way. Furthermore, Frank diSilvestro's three image demonstration above I think answers the question, in the sense that he clearly shows the superiority of just processing a raw file in Lr.

[Ray Harrison]

Of course I am, and a few previous ones are on my Flickr site:

https://www.flickr.com/photos/20926615@N05/albums/72157678279709128

I guess I am not clear though on what you find better than before from looking at the images. What’s a good before and after image?

Sharpening and noise handling are available elsewhere, of course, and presumably if you are shooting jpeg, you’ve got your camera set to optimal jpeg output for the situation and include some level of sharpening, for example. When I tried this application on a couple of images that were challenging (shadow detail needed bringing out) I found it was over sharpened and the colors were off. If I had to answer the question of “does the application produce a better way to edit a jpeg than other ways” my answer would be “no” for my use cases. It doesn’t obviously produce raw or really, even a “raw-like experience”. Raw shouldn’t be part of its name but I think we all agree to that. It does produce a sharper seeming image if your images aren’t sharp, though it has to be dialed down in my case. I find it’s color treatment off, perhaps pilot error on my part though. It quite obviously isn’t going to be able to add details in shadows that I could pull using a standard raw processing approach. I obviously won’t be rushing out to buy it since it adds no real value to my workflow. I respect that others find value.

I will try and post image results later this weekend.