Photo Corners headlinesarchivemikepasini.com


A   S C R A P B O O K   O F   S O L U T I O N S   F O R   T H E   P H O T O G R A P H E R

Enhancing the enjoyment of taking pictures with news that matters, features that entertain and images that delight. Published frequently.

Test Drive: Topaz Labs JPEG to Raw AI Share This on LinkedIn   Share This on Google   Tweet This   Forward This

31 January 2019

We know, we know, we know. There's no such thing as a free lunch. But we accept every invitation anyway. So when Topaz Labs announced JPEG to Raw AI, its tool to convert a simple JPEG image into an editable Raw file, we tucked a napkin under our chin.

Topaz Labs JPEG to Raw AI. Basic interface with Preview enabled.

After downloading the stand-alone application and installing it we loaded it up with a set of recent iPhone 6 Plus JPEGs we knew were high contrast, low-light images that needed some work.

While the iPhone does its best to tweak the captured data before writing the JPEG, we always edit the iPhone images we publish using Photoshop's Camera Raw Filter. At a minimum, we significantly bump on the Clarity or microcontrast, open up the Shadows and bring back the Highlights.

And it's always a dramatic improvement.

So we wondered not so much if JPEG to Raw AI (henceforth J2R) could make an editable Raw file and ultimately a better JPEG image as we wondered how it compared to Photoshop's Camera Raw Filter, which just treats your JPEG to all the processing power of Adobe's Camera Raw.

RUNNING J2R

J2R is designed to be run in batch mode but our system has never quite strained quite so much doing so.

On average it took 17 seconds to convert each iPhone JPEG. And the original 2.0 to 2.8-MB file size ballooned to 48.2-MB. Our camera-generated images of 4320x2868 pixels are only about 14-MB compared to the 3264x2448-pixel image size of the iPhone images.

Graphics Information. As reported by J2R.

J2R will use your GPU. And it seemed to recognize ours, as the Graphics Information report above shows. But it is far short of the minimum 2-GB GPU VRAM Topaz minimum although it is OpenGL 3.3 aware.

On the plus side, J2R writes a DNG or TIFF file. We like to work with DNGs.

Fans running, we never did get through the whole set of images because even though J2R was set to run in the background, our entire system slowed to a crawl making it difficult to swipe from one desktop to another (we keep a set of them cooking) or even to type text.

In addition to running multiple desktops, we run multiple computers so we could have dedicated one machine to the task but we were curious about our comparison, so we aborted the batch to take a look at the results.

CONCEPT

It never hurts to understand what a tool is intended to do. And Topaz Labs has been unusually public about J2R. Albert Yang wrote Can JPEGs Be Improved to RAW Quality? for PetaPixel just a couple of days ago. And the entire staff did an Ask Me Anything on Reddit about the same time.

Topaz Team. Left to right: Albert, Russell, John, Celeste, Eric, Partha.

In addition, the home pages for J2R sums things up nicely. J2R addresses these aspects of your image file:

  • Dynamic Range
  • Color Space
  • Color Depth
  • Compression Artifacts
  • Detail Level

In his PetaPixel piece, Yang explained the approach J2R takes to enhancing those aspects of the JPEG image:

Instead of trying to reverse the information loss, which is impossible, we train a neural network to "remember" what the Raw image should look like before it's converted to a JPEG image. We train the neural network on a large number of high-quality Raw images together with their corresponding JPEG version so that it learns and remembers the correspondence. After training the network on enough images, the neural network will "remember" to output a higher quality image even when given a new "unseen" JPEG image as input. It's just like when you see a friend from far away. Though they are too far away for you to make out the details with your eyes, you can imagine or remember what your friend looks like in vivid detail.

Or, as he summed it up in the Reddit AMA, "For the current release, it does great to recover details, remove artifacts and convert to 16 bit."

Color space is another issue, one the team is working on for the next version, but you can always adjust color temperature in a Raw edit.

And J2R will convert that 8-bit JPEG into 16-bit data so you have some headroom.

While there are settings for blur and noise, J2R isn't primarily concerned with them. Topaz Labs makes another product for that: AI Clear.

It's also important to note that the file J2R produces is not optimized. For that, you have to edit it yourself. J2R makes it possible to edit the file with a great more latitude than the JPEG.

But, we wondered, does it perform better than just using Photoshop's Camera Raw Filter?

COMPARISON

We selected an iPhone JPEG that had plugged up shadows and blown highlights and ran it through Photoshop's Camera Raw Filter and J2R. We edited the J2R file with Photoshop's Camera Raw Filter, as well.

We were surprised that we made almost identical slider shifts for both images with the Camera Raw Filter.

In the full-size crops below you can get a sense of the improvements both edits made to the original image:

In this particular image, we were able to recover quite a bit of shadow detail (the dark coats) although the burned out highlights (lamps) didn't yield any more information.

The rollover above is a 100 percent crop of the full images.

The main difference is a shift in the midtones in which the J2R edit is a bit lighter than the Photoshop edit. You can attribute that to our edit, not the capability of either approach. In the Photoshop edit we were keen on getting all the detail in the central image's dark coat.

But in both edits note the improvement in the details of the black coat in the foreground. And also note the lamps did not yield any more highlight detail.

In fact, in the 100 percent crop you'll notice a bit more "detail" in the shadows (particularly in the gray pants). Is that detail or enhanced microcontrast (or artifacts)? We'd prefer microcontrast, frankly, rather than invention.

CONCLUSION

We won't penalize J2R for hogging our laptop's resources (given its age) and we do applaud its rewriting of JPEG image data to a Raw format that can be edited, although we find the file size dispiriting to put it mildly.

We have long argued for editing iPhone JPEGs with Photoshop's Camera Raw Filter (or the Lightroom equivalent) so we also applaud J2R for making it possible to edit a JPEG in other Raw processors. You really don't know what you're missing until you start sliding those sliders.

The results compared to the Photoshop edit were not worth the overhead of file processing or file size for us, so we're given it three instead of four corners.

But three corners are enough to suggest downloading the 30-day demo and giving it a spin. We certainly plan to continue our own evaluation with other images.

There may be no such thing as a free lunch but sometimes you can just happen to find some free bar food.


BackBack to Photo Corners