On 09/15/2009 03:50 AM, Jenny wrote:

Hi all,

In this Google Summer of Code season, a new detail refinement brush was done. This page is a demo: http://sites.google.com/site/gsoc2009/result-demo

There might still be some bug. I'd love to get any feedback from you.

Hi Jenny,

Thanks for making that page. Would it be possible to also get a sample how the new SIOX performs for green screens? A zoom in at the edge of the extracted object to show how it handles alpha would also be interesting.

These are the kind of pictures I mean: http://images.google.com/images?q=green+screen

BR, Martin

Hi Martin,

Conceptually, these pictures should work very well with SIOX -- also with the version that is already productively included in GIMP. Of course, individual cases might sometimes be more tricky. What Jenny added is the capability of a soft segmentation. This means segmenting regions where background and foreground might fall into the same pixel because of texture complexity or blurring of the picture.

Gerald

-- Dr. Gerald Friedland
International Computer Science Institute 1947 Center Street, Suite 600
CA-94704 Berkeley, USA
http://www.gerald-friedland.org
--

On Mon, Sep 14, 2009 at 9:37 PM, Martin Nordholts wrote:

On 09/15/2009 03:50 AM, Jenny wrote:

Hi all,

In this Google Summer of Code season, a new detail refinement brush was done. This page is a demo: http://sites.google.com/site/gsoc2009/result-demo

There might still be some bug. I'd love to get any feedback from you.

Hi Jenny,

Thanks for making that page. Would it be possible to also get a sample how the new SIOX performs for green screens? A zoom in at the edge of the extracted object to show how it handles alpha would also be interesting.

These are the kind of pictures I mean: http://images.google.com/images?q=green+screen

BR, Martin

--

My GIMP Blog:
http://www.chromecode.com/

Well, every background extraction is tricky - I tried PhotoshopCS 4 tools - they seems to be trivial to use and seem to be easy - but for a complex task which need a lot of pixel precision you have to do a lot of manually corrections too so the final feelig is a kind of frustration - (with gimp magic wand progresive selection feature [drag over layer left / right], I can do the same thing quicker).

In comparison with Photoshop - Jenny demos look perfect especially the tree. I hope the final tool will extract selection / selected colors direct (not just select pixels).

2009/9/15 Gerald Friedland :

Hi Martin,

Conceptually, these pictures should work very well with SIOX -- also with the version that is already productively included in GIMP. Of course, individual cases might sometimes be more tricky. What Jenny added is the capability of a soft segmentation. This means segmenting regions where background and foreground might fall into the same pixel because of texture complexity or blurring of the picture.

Gerald

-- Dr. Gerald Friedland
International Computer Science Institute 1947 Center Street, Suite 600
CA-94704 Berkeley, USA
http://www.gerald-friedland.org
--

On Mon, Sep 14, 2009 at 9:37 PM, Martin Nordholts wrote:

On 09/15/2009 03:50 AM, Jenny wrote:

Hi all,

In this Google Summer of Code season, a new detail refinement brush was done. This page is a demo: http://sites.google.com/site/gsoc2009/result-demo

There might still be some bug. I'd love to get any feedback from you.

Hi Jenny,

Thanks for making that page. Would it be possible to also get a sample how the new SIOX performs for green screens? A zoom in at the edge of the extracted object to show how it handles alpha would also be interesting.

These are the kind of pictures I mean: http://images.google.com/images?q=green+screen

BR, Martin

--

My GIMP Blog:
http://www.chromecode.com/

_______________________________________________ Gimp-developer mailing list
Gimp-developer@lists.XCF.Berkeley.EDU https://lists.XCF.Berkeley.EDU/mailman/listinfo/gimp-developer

_______________________________________________ Gimp-developer mailing list
Gimp-developer@lists.XCF.Berkeley.EDU https://lists.XCF.Berkeley.EDU/mailman/listinfo/gimp-developer

On Fri, Sep 18, 2009 at 2:34 AM, SorinN wrote:

Well, every background extraction is tricky - I tried PhotoshopCS 4 tools - they seems to be trivial to use and seem to be easy - but for a complex task which need a lot of pixel precision you have to do a lot of manually corrections too so the final feelig is a kind of frustration - (with gimp magic wand progresive selection feature [drag over layer left / right], I can do the same thing quicker).

Are you talking about
http://help.adobe.com/en_US/Photoshop/11.0/WSFD9BA8C5-31BA-4fec-81F3-CF04EE5295FCa.html ?

Alexandre

Hi Alex,

Background extraction IS indeed tricky.

First, different pictures require different tools. Everything where the foreground color is essentially one color, such as drawings will work best with a tool like Magic Wand. The foreground extraction Jenny was improving is intended to be used on photographs and works best when the fotograph features a clearly distinctive foreground but the foreground can easily contain millions of colors and as of now, the foreground can also have very fine structure. There is virtually no tool that can deal with transparencies, reflections, and other nasty stuff. When extracting objects with these issues you have to be lucky.

Second, the way to think of these semi-automatic extraction tools is to compare them with a dish washer. Very often, the dish washer will do a good job and clean your dishes. So it'll save you work and it'll be cleaner than if you'd done it manually in the same time. However, for some pieces, the dish washer just doesn't work. Often these are the pieces that are particularly difficult, sometimes though you ask yourself: Why is this glass still dirty -- it's like all the other glasses? So there are people who do not want a dish washer because they want to be in absolute control of the cleaning process. However, would you stop producing, selling, using, and improving dish washers in general, just because they don't work always? The answer is of course: No because in sum they are useful. Same with automatic foreground extraction methods: For some images they save a lot of work, for others they might cause trouble. Some people will never use the tools because they want to be in complete control of the segmentation process. In sum they are useful though.

Gerald

On Thu, Sep 17, 2009 at 10:57 PM, Alexandre Prokoudine wrote:

On Fri, Sep 18, 2009 at 2:34 AM, SorinN wrote:

Well, every background extraction is tricky - I tried PhotoshopCS 4 tools - they seems to be trivial to use and seem to be easy - but for a complex task which need a lot of pixel precision you have to do a lot of manually corrections too so the final feelig is a kind of frustration - (with gimp magic wand progresive selection feature [drag over layer left / right], I can do the same thing quicker).

Are you talking about
http://help.adobe.com/en_US/Photoshop/11.0/WSFD9BA8C5-31BA-4fec-81F3-CF04EE5295FCa.html ?

Alexandre

But probably - if we can try to identify some generic use cases and then to identify a sum of possible techniques / technologies to solve those different cases, we can put a base for a future meta tool.

GIMP has already some useful tools such as color to alpha, and color erase (for brushes), also Gmic has color replace. Maybe if we can have the posibility to pick (with a picker) which color is important to remain and then to pick which color (or range of colors) can go to aplha (probably with a color tolerance control [based on luminance, or other factors]), we can have a better precision for this (meta)tool, and saving a lot of time.

This can go well in SIOUX tool. The same tool as is now but with some [+] color and [-] color selectors / pickers - which will manually refine the alghorithm after the initial selection is done (as is now in SIOUX). When the color selection manually refined is ready, our SIOUX based tool will know much better (if not exactly) about our intention, about which color is important and which is not.

2009/9/19 Gerald Friedland :

Hi Alex,

Background extraction IS indeed tricky.

First, different pictures require different tools. Everything where the foreground color is essentially one color, such as drawings will work best with a tool like Magic Wand. The foreground extraction Jenny was improving is intended to be used on photographs and works best when the fotograph features a clearly distinctive foreground but the foreground can easily contain millions of colors and as of now, the foreground can also have very fine structure. There is virtually no tool that can deal with transparencies, reflections, and other nasty stuff. When extracting objects with these issues you have to be lucky.

Second, the way to think of these semi-automatic extraction tools is to compare them with a dish washer. Very often, the dish washer will do a good job and clean your dishes. So it'll save you work and it'll be cleaner than if you'd done it manually in the same time. However, for some pieces, the dish washer just doesn't work. Often these are the pieces that are particularly difficult, sometimes though you ask yourself: Why is this glass still dirty -- it's like all the other glasses? So there are people who do not want a dish washer because they want to be in absolute control of the cleaning process. However, would you stop producing, selling, using, and improving dish washers in general, just because they don't work always? The answer is of course: No because in sum they are useful. Same with automatic foreground extraction methods: For some images they save a lot of work, for others they might cause trouble. Some people will never use the tools because they want to be in complete control of the segmentation process. In sum they are useful though.

Gerald

On Thu, Sep 17, 2009 at 10:57 PM, Alexandre Prokoudine wrote:

On Fri, Sep 18, 2009 at 2:34 AM, SorinN wrote:

Well, every background extraction is tricky - I tried PhotoshopCS 4 tools - they seems to be trivial to use and seem to be easy - but for a complex task which need a lot of pixel precision you have to do a lot of manually corrections too so the final feelig is a kind of frustration - (with gimp magic wand progresive selection feature [drag over layer left / right], I can do the same thing quicker).

Are you talking about
http://help.adobe.com/en_US/Photoshop/11.0/WSFD9BA8C5-31BA-4fec-81F3-CF04EE5295FCa.html ?

Alexandre
_______________________________________________ Gimp-developer mailing list
Gimp-developer@lists.XCF.Berkeley.EDU https://lists.XCF.Berkeley.EDU/mailman/listinfo/gimp-developer

--
Dr. Gerald Friedland
International Computer Science Institute 1947 Center Street, Suite 600
CA-94704 Berkeley, USA
http://www.gerald-friedland.org
_______________________________________________ Gimp-developer mailing list
Gimp-developer@lists.XCF.Berkeley.EDU https://lists.XCF.Berkeley.EDU/mailman/listinfo/gimp-developer

Hi,

That's a great idea. The idea would be to have only on foreground selection tool and make it automatically determine whether to use Magic Wand or SIOX (btw.: not spelled like the native American tribe). This shouldn't be very hard since Magic Wand's user interaction is a subset of SIOX user interface. The determination whether to use one or the other could depend on the amount of colors that are found in the user-defined foreground sample. There might be a radio-button to override the automatic choice.

Any volunteer? Maybe it's for Google Summer of Code 2010...

Gerald

On Fri, Sep 18, 2009 at 5:02 PM, SorinN wrote:

But probably - if we can try to identify some generic use cases and then to identify a sum of possible techniques / technologies to solve those different cases, we can put a base for a future meta tool.

GIMP has already some useful tools such as color to alpha, and color erase (for brushes), also Gmic has color replace. Maybe if we can have the posibility to  pick (with a picker) which color is important to remain and then to pick which color (or range of colors) can go to aplha (probably with a color tolerance control [based on luminance, or other factors]), we can have a better precision for this (meta)tool, and saving a lot of time.

This can go well in SIOUX tool. The same tool as is now but with some [+] color and [-] color  selectors / pickers - which will manually refine the alghorithm after the initial selection is done (as is now in SIOUX). When the color selection manually refined is ready, our SIOUX based tool will know much better (if not exactly) about our intention, about which color is important and which is not.

2009/9/19 Gerald Friedland :

Hi Alex,

Background extraction IS indeed tricky.

First, different pictures require different tools. Everything where the foreground color is essentially one color, such as drawings will work best with a tool like Magic Wand. The foreground extraction Jenny was improving is intended to be used on photographs and works best when the fotograph features a clearly distinctive foreground but the foreground can easily contain millions of colors and as of now, the foreground can also have very fine structure. There is virtually no tool that can deal with transparencies, reflections, and other nasty stuff. When extracting objects with these issues you have to be lucky.

Second, the way to think of these semi-automatic extraction tools is to compare them with a dish washer. Very often, the dish washer will do a good job and clean your dishes. So it'll save you work and it'll be cleaner than if you'd done it manually in the same time. However, for some pieces, the dish washer just doesn't work. Often these are the pieces that are particularly difficult, sometimes though you ask yourself: Why is this glass still dirty -- it's like all the other glasses? So there are people who do not want a dish washer because they want to be in absolute control of the cleaning process. However, would you stop producing, selling, using, and improving dish washers in general, just because they don't work always? The answer is of course: No because in sum they are useful. Same with automatic foreground extraction methods: For some images they save a lot of work, for others they might cause trouble. Some people will never use the tools because they want to be in complete control of the segmentation process. In sum they are useful though.

Gerald

On Thu, Sep 17, 2009 at 10:57 PM, Alexandre Prokoudine wrote:

On Fri, Sep 18, 2009 at 2:34 AM, SorinN wrote:

Well, every background extraction is tricky - I tried PhotoshopCS 4 tools - they seems to be trivial to use and seem to be easy - but for a complex task which need a lot of pixel precision you have to do a lot of manually corrections too so the final feelig is a kind of frustration - (with gimp magic wand progresive selection feature [drag over layer left / right], I can do the same thing quicker).

Are you talking about
http://help.adobe.com/en_US/Photoshop/11.0/WSFD9BA8C5-31BA-4fec-81F3-CF04EE5295FCa.html ?

Alexandre

Gerald Friedland wrote:

That's a great idea. The idea would be to have only on foreground selection tool and make it automatically determine whether to use Magic Wand or SIOX

I have been saying this for a while:

every time I look at the 4 'content' selection tools (wand, scissors, SIOX
and by-color) I think: that should just be one tool with some options, or
even a mixer to get the balance of results of different algo's.

it seems that there is a ramp-up in sophistication in the wand -> by-color -> SIOX order. scissors seems to be and handle quite different. but after discussing its usefulness today, we can add those 'different' components to the mix.

--ps

founder + principal interaction architect man + machine interface works

http://mmiworks.net/blog : on interaction architecture

On 09/19/2009 03:17 PM, peter sikking wrote:

Gerald Friedland wrote:

That's a great idea. The idea would be to have only on foreground selection tool and make it automatically determine whether to use Magic Wand or SIOX

I have been saying this for a while:

every time I look at the 4 'content' selection tools (wand, scissors, SIOX
and by-color) I think: that should just be one tool with some options, or
even a mixer to get the balance of results of different algo's.

But SIOX is supposed to be more than a selection tool, isn't it? Isn't it about extracting foreground objects? A selection is not enough for that when the foreground consists of transparent pixels.

/ Martin

On Sat, 2009-09-19 at 15:38 +0200, Martin Nordholts wrote:

But SIOX is supposed to be more than a selection tool, isn't it? Isn't it about extracting foreground objects? A selection is not enough for that when the foreground consists of transparent pixels.

SIOX is a tool to create a selection, nothing more. I don't see what a foreground extraction tool should do besides selecting the foreground. Perhaps you can explain this?

Sven

On 09/19/2009 08:20 PM, Sven Neumann wrote:

On Sat, 2009-09-19 at 15:38 +0200, Martin Nordholts wrote:

But SIOX is supposed to be more than a selection tool, isn't it? Isn't it about extracting foreground objects? A selection is not enough for that when the foreground consists of transparent pixels.

SIOX is a tool to create a selection, nothing more. I don't see what a foreground extraction tool should do besides selecting the foreground. Perhaps you can explain this?

Consider a completely red background with a slightly blurred green filled circle in the foreground [1]. If you want to extract the green foreground object you need to get rid of the red in the pixels with both red and green, but you can't do that with only a selection.

A good foreground selection tool would need to get rid of the red in the pixels with both red and green and only leave slightly transparent green pixels.

/ Martin

[1] http://www.chromecode.com/temp/red-circle-on-green-background.png

Hi,

But SIOX is supposed to be more than a selection tool, isn't it? Isn't it about extracting foreground objects? A selection is not enough for that when the foreground consists of transparent pixels.

SIOX is a tool to create a selection, nothing more. I don't see what a foreground extraction tool should do besides selecting the foreground. Perhaps you can explain this?

Consider a completely red background with a slightly blurred green filled circle in the foreground [1]. If you want to extract the green foreground object you need to get rid of the red in the pixels with both red and green, but you can't do that with only a selection.

A good foreground selection tool would need to get rid of the red in the pixels with both red and green and only leave slightly transparent green pixels.

Absolutely. And this is exactly what Jenny has implemented as part of the Google Summer of Code.
(So until now SIOX was 'only a selection tool' but now it's not anymore -- at least once Jenny's code makes it's way into the main branch)

Gerald

-- Dr. Gerald Friedland
International Computer Science Institute 1947 Center Street, Suite 600
CA-94704 Berkeley, USA
http://www.gerald-friedland.org

On 09/20/2009 08:19 AM, Gerald Friedland wrote:

A good foreground selection tool would need to get rid of the red in the pixels with both red and green and only leave slightly transparent green pixels.

Absolutely. And this is exactly what Jenny has implemented as part of the Google Summer of Code.
(So until now SIOX was 'only a selection tool' but now it's not anymore -- at least once Jenny's code makes it's way into the main branch)

Ok cool.

Speaking of integrating into git master, who will do that? There is a need to cleanup the branch and introduce proper commit messages. If we could get help with that it would be great.

BR, Martin

Hi

A good foreground selection tool would need to get rid of the red in the pixels with both red and green and only leave slightly transparent green pixels.

Absolutely. And this is exactly what Jenny has implemented as part of the Google Summer of Code.
(So until now SIOX was 'only a selection tool' but now it's not anymore -- at least once Jenny's code makes it's way into the main branch)

Ok cool.

Speaking of integrating into git master, who will do that? There is a need to cleanup the branch and introduce proper commit messages. If we could get help with that it would be great.

I think the integration should be a collaborative effort between Jenny and somebody who actually knows the GIMP code very well and has commit privileges on the main tree. I am happy to help conceptually, meaning on advising how the output should look like and what is possible and what might not be possible for further extensions.

Gerald

-- Dr. Gerald Friedland
International Computer Science Institute 1947 Center Street, Suite 600
CA-94704 Berkeley, USA
http://www.gerald-friedland.org
--
Sent from Berkeley, CA, United States