Archive for the ‘Uncategorized’ Category

Rendering Synthetic Objects into Legacy Photographs

October 30, 2011

Just a quick note:

Kevin Karsch posted a video on vimeo under the title:

“Rendering Synthetic Objects into Legacy Photographs”
at Rendering Synthetic Objects into Legacy Photographs which was published at SIGGARPH Asia 2011
He (/his group ?) takes a 2D picture and users mark 3D-axis-system in the picture and some further objects in the scene. Moreover they mark light sources. From this, he can add objects in the scene which are lit more realisticly then without knowing the light sources. He did a user study on the quality of his results.
(I found this via StumbleUpon on howtoweb.com and I’m still searching for a good site to get updates on research like this. (Okay, maybe I should check out the SIGGRAPH papers more thoroughly ))

Content Aware Fill in Photoshop and Gimp’s Resynthesizer Tool

March 29, 2010

Just a short note about Content Aware Fill. You might have seen the youtube video by Adobe and maybe you didn’t know that this was possible.

Here’s the video and the feature will be available in the next version of Photoshop

Turns out, however, that GIMP already supports this feature via a plugin. Maybe it’s not 100% as easy to use, but non the less it’s in the reach for everybody to try. Here is the video:

(There is another video at GIMP plugin “Resynthesizer” and “Single Window” mode that gets the same result on the last picture by just resynthesizing on selection.)

I think one can really only compare the two products when looking at high resolution pictures, so a final say has to await the release of CS5.

But you have to hand it to Adobe for their PR skills. I really like their video, it impressed me, and there had been a good chance that I hadn’t found out about alternatives already available to me.

And then one more: There is this video showing some artifacts created by GIMP Resynthesizer and how to fix those. (unfortunately the audio is a little to fast for the video.)

Google research takes 40mil pictures, finds landmarks, matches against your vacation photos

March 28, 2010

Mid 2009, Google showed off a research paper at CVPR 2009. You can read about it on Google’s blog: A new landmark in computer vision.

What they did:

Google research took 40mil pictures, several of them geotagged. Then they  automatically organized them into clusters per landmark (unsupervised). And afterwards they are able to take an arbitrary picture of a landmark (for example from your last vacation) and tell you with a 80% chance what the name of the landmark is and what the location is.

From their 8-page research paper:

The GPS-tagged photos yield ! 140k geo-clusters and !14k visual clusters with text tags, from which 2240 landmarks from 812 cities in 104 countries are mined.

I’ve read about it first on popsci.com. And I’m still searching for a blog that would have told me earlier. Read about what blog I’m searching for in my first post: Is there a blog on cool image processing stuff for geek end-users?

Making “Building 3D models” into a game

March 28, 2010

You might already know Google Building Maker, which is a little like a game. But here comes PhotoCity from the University of Washington and puts the “claiming a portion of the real world and get a badge for it” into the space of making 3D models.

There are photos of a location and program like PhotoSynth makes a 3D building from that. Now the program defines spaces in the neighborhood of the 3D model. It’s your task to take photos of these spaces (overlapping with known parts of the existing 3D-model), so that the 3D model can become larger. You can then claim the region that was added because of your pictures.

Here is their intro video:

But they don’t have a region near my home just yet. It says on the homepage that one can seed new regions but I don’t know how that works and who is allowed to do that. It seems all it takes is to register to the site and I will update this blog, when I have done that.

Photosynth in faster

March 28, 2010

There is a article on popsci.com that states that the research on the code going into PhotoSynth has gone on:

The grail section of the computer science department of the University of Washington has released new papers in 2009 that describe how one can match more photos in a PhotoSynth like way, in order to create 3D models of objects (houses / town).

You can find their site at: Building Rome in a Day.

There are some examples on youtube:

Content-aware image resizing (CAIR, seam carving)

March 28, 2010

Hello, this i very old news, but I wanted to add this topic to the post anyhow, so you know what kind of content I plan on including here (till I find another blog that does the curating for me).

For now I let the video speak for itself:

You can do this in Photoshop from CS4 as content-aware scaling, there is a GIMP plugin, there is implementation on Source Forge and you can get a google-code GUI for that.

I have downloaded those but I haven’t tried them.

I have however used a flash version of CAIR on http://rsizr.com/.  It seems to me, that the whole algorithm is implemented into the flash file, so I that all the calculation is done on your PC and not on a server. I don’t know whether the owner of the site is to be trusted that he won’t keep a copy of the picture, but I just hope he won’t. And one more thing about the flash version. My results were much, much better than the ones Stephan got when he tried it.