Content Aware Fill - Demystification

Nothing sexy, just a very quick test using low res jpegs, it took a couple of minutes:

180842mainflag.jpg
180842mainflagnonejpg.jpg
film80.jpg
film80none.jpg


It’s not a miracle tool but with some work it will do for a quick fix or two.
 
Its amusing people think that tick box is going to perform some kind of miracle like on tv shows where they magnify the reflection of a killer in a shop window on a cctv photo and suddenly have a pixel perfect picture of the bad guy. Do you think Scotland Yard have upgraded to CS5 yet?
 
The thing is, it clearly is capable of doing what was shown in the videos produced by Adobe. However, one must remember (as noted above in this thread) that it will only work in some circumstances, such as random repeated patterns (such as grass, sky etc). Of course, Adobe carefully selected the image to be used in the video (and why wouldn't they?).

Conceptually, I still think its a great tool -we all just need to accept that marketing is marketing, and the reality across a spectrum of photos will be slightly different.
 
.... and a quick tidy up:

heron1.jpg
heronnone1.jpg

There are quite a few situations where it won’t work but it’s still a very useful addition to the toolkit. :)
 
So can someone explain why Willis' hasnt worked?

Because Dennis and Asim and Simulator are all making decent sized selections. With what I would have thought is less patching data for the algorithm. The sky should be an easy one for it to handle and yet its using the clouds as the patch?
 
gord said:
So can someone explain why Willis' hasnt worked?

Because Dennis and Asim and Simulator are all making decent sized selections. With what I would have thought is less patching data for the algorithm. The sky should be an easy one for it to handle and yet its using the clouds as the patch?
Click to expand...

Without knowing how it goes about applying the task I have no idea.
I have just tried to remove the clouds using multipue lasso selections and I have got rid of the entire cloud but am left now with a slightly off tonal patch in half of the area.

So far I have most of my best sucess with heavily patterned areas rather than gradient areas.
 
asim18 said:
You should use the healing brush for that sort of work. Took me about 30 seconds.

kbigil1b_a18.jpg
Click to expand...

LOL yeh I do know that but I was just showing people a simple action that the Filter could not do. like people have said I think it a great tool and used with other tools will save on time but don't rely on it to sort everything out. ;)
 
greedy123 said:
Does the content aware work on 16bit images as well as 8 bit?
Click to expand...

On the subject of which, CS5 now allows you to save an 8bit jpeg when working on a 16bit file without having to convert down before saving. It just performs the conversion at output. Which is just ruddy marvellous.
 
gord said:
So can someone explain why Willis' hasnt worked?

Because Dennis and Asim and Simulator are all making decent sized selections. With what I would have thought is less patching data for the algorithm. The sky should be an easy one for it to handle and yet its using the clouds as the patch?
Click to expand...

From my experiments with the Content Aware Tools, it seems to me that when using content aware with the healing brush the the size of the area it resamples is determined, at least in part, by the diameter of the brush being used.

I've noticed that if I have a fairly large area that you want to remove, if I use the healing brush and choose a large brush to quickly cover the area, it seems to sample from a big area. If there are distinct things in the surrounding area, they do tend to get sampled, like in Willis' example. If I then tried again with a much smaller brush, ok it took longer to select but the sampling area seemed to be smaller and I was much less likely to end up with effects like in Willis' example.

As for the fill, I think in the example, as others have said the problem lies with the selection that has been made. If it were closer to the area to be removed it would not be sampling quite so close to the elements that have ended up in the filled area.
 
Mr Jones said:
For me content awareness is useful, for some pictures, it's down right impressive for a 10 second jobbie.
http://www.flickr.com/photos/spyker3292/4566795145/
On some other images though, it's great for the first pass then touch up.
Click to expand...

Exactly right, some photos it's brilliant others it gives you a great first effort ready for manual touching up.

That video shows exactly how good content aware can be!! And is also a good example of using it in the correct circumstances, ie a fairly random patterned background.

As with any tool in Photoshop, it takes time to learn exactly what the tool can do and then it takes good judgement by the user as to which tool is correct for the job in hand!
 
Last edited:
valve90210 said:
Exactly right, some photos it's brilliant others it gives you a great first effort ready for manual touching up.

That video shows exactly how good content aware can be!! And is also a good example of using it in the correct circumstances, ie a fairly random patterned background.

As with any tool in Photoshop, it takes time to learn exactly what the tool can do and then it takes good judgement by the user as to which tool is correct for the job in hand!
Click to expand...

Thaaank you. Exactly what I have been trying to say from the beginning.
 
You must log in or register to reply here.
Back
Top Bottom