Improvements to this release:
- You can mark areas of the image that you would like to retain/remove with a paint style mouse gesture.
- JPG support (built into the windows binary)
- I am now using the Qt Graphics View Framework, which should make image handing (zooming etc) a bit snappier.
- Update: I got notified of a few bugs, so I fixed those using valgrind
- Update: While I was at it, I implemented copy and paste in windows
- Various algorithm speed improvements
- Copy and Paste [Update: completed in latest build]
- Anything else you guys suggest
Again the download links are
- Update: All Downloads Now at the seam-carving-gui project Download Page
Here are some images showing the program in action...
A nice castle image before any retargeting:
Mark areas that we want to remove and we want retain, in this case we want to make sure the castle is not distorted and remove the pesky pedestrian:
The re targeted image, pretty slick eh?
Update: The latest version is no longer based on Andy's implementation but CAIR. See the seam-carving-gui project page for more.
6 comments:
Nice example, but maybe you should cut the person's shadow out as well!
There's a flickr group for this kind of thing now, with links to your software (which I've used, cool)
http://www.flickr.com/groups/452426@N25/
Very cool initiative. I hope you will make the version for OS X too! I don't know enough about programming to do...
Very cool! Also stumbled it ;) That should increase your traffic.
Hi Gabe. I've been working on a implementation of the resizer. I believe it to be a close representation of the paper, although I haven't looked at your code yet to compare. I made some comparison images of our two methods, and mine does produce different results, with some less artifacts. However, in its current form its quite slow. Email me at brain.recall@gmail.com and I'll send you the examples. I still got a little work to do on my code yet.
Gabe, the program works pretty good, but there is one thing that makes it tough to use - the fact that one has to guess the right size of the smaller picture. If my guesstimate is poor, I'll have to apply the algorithm again, marking the red and the green zones, then making yet another guess.
Perhaps it would be better if:
- I could press undo, and choose a smaller image size
- I could decrease the image size incrementally, while the program memorizes the weights, so I don't have to keep re-assigning areas.
Where's the makefile and lib dependencies?
Ah, the website had it if I looked harder. But the resize operation is definitely not right.
Post a Comment