I'm unable to read the second paper due to not being an ACM member.... maybe someone who has read it can post the gist of it here? Otherwise, I suppose I'll hear about it in class.
I found DeCarlo and Santella to be an interesting read, but somewhat unsatisfying in that the authors seemed too concerned with making pretty results. While that is a fine thing for an NPR system to do (and I find their results to be quite good), I would have liked to see more empirical evaluation. In particular, the authors somehow overlooked an important, and seemingly obvious (to me at least) sanity check, namely: are fixations in the original a good prediction of fixations in the result?
I think the answer to that question might speak to the usefulness of their method more so than qualitatively analyzing their pictures. This seems especially true since the technique doesn't seem very practical--at least until eye trackers become cheaper and easier to use, or computer vision techniques can predict fixations more accurately.
I agree that the DeCarlo paper could have benefited from some more quantitative results (though the pictures are certainly pretty). The other paper does do several things to measure the value of their automatic crops which could also have been applied to the DeCarlo work. Foremost is the idea of performing user preference studies- I would imagine people will prefer the gaze-based abstraction, but it's hard to tell.
The other thing the cropping paper does that would have been interesting to see in the stylization paper is the comparison against automated saliency methods as well as manual artistic saliency marking.
I'm unable to read the second paper due to not being an ACM member.... maybe someone who has read it can post the gist of it here? Otherwise, I suppose I'll hear about it in class.
ReplyDeleteI found DeCarlo and Santella to be an interesting read, but somewhat unsatisfying in that the authors seemed too concerned with making pretty results. While that is a fine thing for an NPR system to do (and I find their results to be quite good), I would have liked to see more empirical evaluation. In particular, the authors somehow overlooked an important, and seemingly obvious (to me at least) sanity check, namely: are fixations in the original a good prediction of fixations in the result?
I think the answer to that question might speak to the usefulness of their method more so than qualitatively analyzing their pictures. This seems especially true since the technique doesn't seem very practical--at least until eye trackers become cheaper and easier to use, or computer vision techniques can predict fixations more accurately.
I agree that the DeCarlo paper could have benefited from some more quantitative results (though the pictures are certainly pretty). The other paper does do several things to measure the value of their automatic crops which could also have been applied to the DeCarlo work. Foremost is the idea of performing user preference studies- I would imagine people will prefer the gaze-based abstraction, but it's hard to tell.
ReplyDeleteThe other thing the cropping paper does that would have been interesting to see in the stylization paper is the comparison against automated saliency methods as well as manual artistic saliency marking.