I may have missed this, but were any performance numbers given for the finding an object in a new image? Also, even though this seems to work pretty well when the number of training examples is low, in your opinion is it a big problem that the training complexity goes up quadratically with the number of training images?

This was exactly what we were trying to do with ReSPEC. Finding consistency in an unlabeled set of images, and extracting that consistency (object). I have no real comment other than to say this paper is awesome and I will respectfully let them use the name ReSPEC for their method, so long as we get cited :) LONG LIVE ReSPEC

Ignoring the fact that training takes so long, this seems like the perfect application to learning a map from keywords for an image search query to a segmentation tree that is representative of most of the objects in the set of images.

How about applying this approach to Grozi? We have a grocery list and want to learn a representation for these objects without doing much work. We can feed each grocery item into an image search engine, get the top N results, feed them into this framework, and get out a representation for the grocery item. Make sense?

While the authors point out that their error metric is strict compared to some of Perona's work and that it corresponds well with their own intuitions, it still feels like finding the target area +/-25% is being pretty generous. I'd be interested to see how their error changes as they adjust their thresholds for a true positive.

I think their region shape descriptor (histogram of pie slices) is pretty neat. Also, considering the tiny amount of training data their method requires, their results are pretty incredible!

Could you go a bit into their segmentation algorithm that this paper depends on? Do you know how its results might compare to the other segmentation papers we read?
Also, they mention that segments are more invariant in lighting, orientation, and scale than features. Is this intuition necessarily true?

So, my question is pretty much the same as Adams. The paper/algorithm they use for segmentation can be found at http://ieeexplore.ieee.org/iel1/34/11961/0...6258&isnumber=11961 . and is interesting in and of itself. It seems to be like a watershed but with some notion of forces (I only halfheartedly read it unfortunately so I don't know too much about it). It seems more invariant because it basically does a multiscale method with intensity relativeness (per scale per region). That should provide both decent illumination and scale invariance. Any chance we could get a tiny little comparison between this and either the advanced segmentation paper we touched or a simple algorithm like watershedding?

Do you have any ideas how this can be extended to video? Thank you!

In the recall-precision curves from figure 8, it seems that Leibe et al outperformed Tree Union. However, they state that Leibe had less demanding evaluation criteria. In your opinion, why didn't they perform any evaluations between Tree Union and the other methods using the same evaluation criteria?

Paul Smith Shoes uk store,See discount Paul Smith large range of online,We have

Paul Smith Shoes and Paul Smith Shoes in all styles.