Adobe admits using 'synthetic blur' image in deblur demo

Adobe has admitted an image used in its 'image deblur' presentation was artificially blurred for the purposes of the demonstration. The company said the blur on the image was 'more complicated than anything we can simulate using Photoshop's blur capabilities.' It described the move as 'common practice in research' and defended the use of the image because 'we wanted it to be entertaining and relevant to the audience.' The other images shown were the result of camera shake, it said.

Source


If you would like to see the video demonstration go to this thread

I suppose this news article brings up something that seems to come a lot on ATS and image manip threads ...

Adobe states that the application didn't 'know' that the blur was synthetic and therefore stands by its example ... however a synthetic blur is a linear function in many cases meaning quite easy to reverse in comparison to say a non-linear random function.

The point spread function which is required to reverse the blur really isn't as difficult to work out as a real world chaotic situation with poor focus, motion blurs, bad lighting, noise, lens flares and all the rest of it. This isn't the first time a deblurring app has reversed a synthetic blur and claimed magical results. It won't be the last. (PS Armap was correct about some of this in a previous thread )

It's just not popular to state the limitations of such functions (or the fact it's based on some many decades old maths functions), and it's something ATS users need to be aware of when they look at, not only their own imagery, but other persons. When looking at what a piece of image software allegedly does you have to look at it from the perspective of a washing up liquid commercial. Ever had to clean soap scum out a shower? Did it come off in one wipe like the advert?

Here is another example raised from a recent thread: Lucis vs. Unsharp Mask Fight!

In this document Lucis compares their product with an unsharp mask. This is like comparing using bleach with rolling your face across the floor and hoping the stain goes away. The unsharp mask was vogue in around the 1930s - 1940s, it's not exactly a new thing.

And another example: Lucis Competitive Advantages PDF

These examples, at best, are poor, and at worst are blatantly misleading; it's a boxing match where the first fighter gets to choose any opponent. Contrast equalization Lucis style techniques and unsharp masks aren't really mathematically similar. They do similar things in very different ways with very different trade offs, but this would make you believe there's something amazing in this new patented algorithm.

Just to give an example ... I might use an unsharp mask to tighten up an image so I can perhaps get an easier tracking point for science or compositing purposes.

I wouldn't want to use a laplacian because for a particular scene it might bring up too much noise, and a histogram equalization technique may also be prone to noise and can turn my blacks to grey. Not only will this throw off my tracking since my track is likely based on luma values, but it will also take more processing power than the unsharp mask and laplacian combined in many scenarios requiring many more calculations per frame. (Though this limitation is rapidly being decreased by bigger computers)

I suppose the point I'm trying to make is ... It's not the tools that make the image analyst/artist, and no special plugin is going to act as a magic bullet regardless of the advertising. If I was against John Knoll in an art competition and Knoll was given two crayons and an elastic band with an out of print copy of photoshop 6.0 emulated onto a rusty old Amiga, and I was allowed to use CS5 and Nuke ... my money might end up being on Knoll. Obviously that's a ridiculous (and maybe optimistic) scenario, but a good, knowledgeable person using software that doesn't cost a million bucks will do well regardless of limitations of their tools.

Image theory doesn't suddenly change because a person spent a large amount of money or bought a plugin. For example, many of the more expensive systems aren't expensive always because they provide new functionality, or have magical tools. There are expensive systems such as Smoke, and Flame in the film industry, and many other expensive systems in the medical and science fields. They often actually lack the functionality of lesser allegedly inferior systems. They have fewer tools developed for them these days because there are many times fewer high end consumers. Generally what you're paying for is things like 24/7 support on call support and reliability, and a single tested one-stop shop system and work flow which can tackle the latest high resolution imagery in real time (often the hardware isn't even physically that cutting edge); in the case of medical imaging you may be recieving some insurance regarding results and assurance that the algorithms and tools provided are accepted scientific standard (therefore often not cutting edge in many cases). In a case where a break in your work flow might cost you a million dollar a year VFX client or, worse, result in someone's death you want tried and tested promise work flow and results.

These big gun apps and hardware don't make a person any better at controlling an image, editing it, or analyzing it. (Quite the opposite in some cases! You will be amazed what happens when a person is allowed to add as many nodes/filters as they want without punishing them by crashing their workspace.) A martial artist learns the very base techniques of their art prior to learning how to Chuck Norris a person in the face. It is the same for image analysts.

Just an example of the price of Smoke which is a video editing system in the autodesk family:

An Autodesk Smoke 2012 for Mac OS X license is available at a suggested retail price of $14,995.* Autodesk Subscription is available for purchase simultaneously with the product license for $1,995 MSRP per year.

Source

Perhaps one day we will have algorithms which not only analyze and process your image, but work out all the steps in between ... so perhaps you may just be able to click the 'I wanna see it better button, kthxgo' ... but until then it's highly likely we're just going to be working on our individual filters and functions and trying to choose the best one for what we're aiming to do.

And until then I really would urge people to learn as much image theory as they can if it is what interests them. It can save you money on expensive ABC plugins, and it's a great investment in the future because, oddly enough, how light, color and maths works isn't going to change in our life times most likely.

Not really surprising seeing that Adobe price gouges all their software. But they can do that because there is no real competition against them. They pretty much have a monopoly on industry standard imaging software. And if there is competition against them they just buy them up, like they did with Macromedia.

reply to post by Pinke


If they want to clear the record then how about another demonstration. We can use some of the NASA photos of blurred items on the moon / mars / etc to see if it works or not. There only saving grace right now is it was just a demonstration with no release, meaning they just barely missed several lawsuits as well as criminal investigation.

reply to post by Xcathdra


The process will not work on blurred photos like NASA's moon pics. It only works on blurring due to motion shake thats caused by camera movement with an incorrect shutter speed.

I do not get why adobe do not release a 3D program to compete with 3Ds Max etc. I really think it is a field they should venture into rather than adding basic features into the likes of photoshop and illustrator.

Originally posted by PhoenixOD
reply to post by Xcathdra

 

The process will not work on blurred photos like NASA's moon pics. It only works on blurring due to motion shake thats caused by camera movement with an incorrect shutter speed.

well that sucks...

oh well

reply to post by OwenGP185


The two markets are completely different. Why combine two independently profitable product lines onto one jack of all trades master of none?

Edit to the above: brain fart moment, autodesk produce 3D Studio Max. Even more of a reason not to try and enter a saturated market against competition who have decades more experience than you.

Originally posted by PhoenixOD
reply to post by Xcathdra

 

The process will not work on blurred photos like NASA's moon pics. It only works on blurring due to motion shake thats caused by camera movement with an incorrect shutter speed.

It will most likely depend on the chracteristics of the blur. If it's fairly linear etc ... then their process may work, but there's generally ways to look at things like that anyway if people really wanted to.

Originally posted by OwenGP185
I do not get why adobe do not release a 3D program to compete with 3Ds Max etc.

Originally posted by john_bmth
reply to post by OwenGP185

 

The two markets are completely different. Why combine two independently profitable product lines onto one jack of all trades master of none?

The two markets actually work together a lot these days. Things like Nuke and After Effects need to deal with 3D objects quite commonly. Though Adobes doesn't have as many experience 3D people as say Autodesk who primarily deal with CAD programs and architecture etc ... though the gap between architecture, image engineering, and VFX work is narrowing quite considerbly.

A lot of it really comes down to people though. IE Avid is one of the 'industry standard non linear editors' over Adobe's and Apple's products. Apple's Shake is an 'industry standard' comp program ... though that's slowly changing. Studios invest large amounts of money into a single line of applications and stay with it for quite sometime. Shake has been discontinued for many years, but continues to limp on a little since studios have invested so much time and training in it ... even though Nuke is faster and better in many ways (and also most importantly still under production!)

The fact of the matter is, the industry is generally a little bit behind since they're quite reluctant to change. Generally I've learnt that the indie crowd are sometimes a little bit ahead of the curve, and the rest of the industry tends to just roll on and pick up the new developments as soon as someone proves how great they are.

Just to give an idea, apps like mud box etc ... were introduced to people like Steven Spielburg by a new person doing design work for them. The production company was impressed how the person could design a creature or space ship in 3D and hand it to the post production team almost ready to use. They were also impressed at how quickly changes can be made ... The apps that artist used are now parts of the work flow used on those bigger films.

It's never about the best or the greatest ... it's about convincing people they're using the best and the greatest. This is unfortunately why these demonstrations are incredibly effective over straight out honesty.

reply to post by Pinke


I don't really see a synergetic benefit for a couple of reasons. Artist-driven solutions like Z Brush are far more intuitive for things like texture generation than marrying something like Max and Photoshop. Also, for the professional market that such a product would be aimed at, texture artist and 3D modeller are two distinct roles. For post-processing, there's no need to be in the modelling software as by it's very nature it's performed in image space. And again, Autodesk have the power of Market leader position as well as a whole collection of ranges, from their modelling, CAD, global illumination and so on. Attempting to crack that market would be an immense challenge.

reply to post by john_bmth


I mostly agree ... however, the new breed of artist is breaking those roles down substantially. Many studios do still have the break down of various roles, but more and more the roles are becoming hybrid. Zbrush is an excellent example of where the design phase is starting to impact the post production phase.

It's the internet's influence mostly. It's not uncommon now to have an artist that's great at 3 or more different fields rather than the old days where you rotoscoped whilst sneaking in after hours moments on a big comp system to learn.