QuickTopic free message boards logo
Skip to Messages


Direct methods for visual scene reconstruction

12:05 PM ET (US)
If you want to Get awareness about bing homepage quiz. Test your knowledge by playing this quiz. https://bingrewards.info/bing-homepage-quiz/
Josh Wills
01:50 PM ET (US)
But aren't those far-off applications a bit "fairytale-ish"?
Kristin BransonPerson was signed in when posted
01:32 PM ET (US)
In reply to Satya: instead of solving for the depth at each pixel in the image, the depth is only solved for at the spline control vertices, which is basically just a subsampling of the image. The depth at each pixel is interpolated from the depths at the control vertices.

In reply to Josh: I think some of the (far off) applications require real-time, for example a tele-football game.
Josh Wills
01:20 PM ET (US)
In section 7, the authors point out that the methods that are based on geometry are correct but there is no guarantee in noisy data. It seems ridiculous to think that this guided search over a very large space of transformations than a method that decomposes the scene into points related by geometric constraints.

Also, it seems that any illumination change or specularity would make it difficult to get a small result for their minimization term. Also they criticize feature-based approaches for inaccuracy of matches but it seems that regions that are difficult to match (like periodic patterns or regions with illumination changes) would also be very difficult in this approach.

One other question that I have is what is the contribution of this paper? They say that they are real-time homography capable, but these applications don't seem like areas that really need real-time
Kristin BransonPerson was signed in when posted
01:16 PM ET (US)
In reply to Sameer: the paper says the sum in Equation (2) is "over all corresponding pairs of pixels i which are inside both images." Of course, there is no mention of how to determine if a pixel is in both images. Perhaps it can be approximated using their hierarchical matching technique if between-frame motion is large enough that a significant number of pixels are not in both images. This is the same idea Serge presented in his Direct Methods vs Feature-Based Methods presentation (Equation (4) of "All About Direct Methods"). It is valid if you assume the brightness constancy constraint.
Edited 11-07-2002 01:17 PM
sameer agarwal
02:38 AM ET (US)
I am disturbed by section 2.1

how can you estimate the homography by simply minimizing the error between the brightness intensities of the original and the transformed image ? Isn't there a mild problem with regularlization here ? in terms of a tradeoff between number of pixels which overlap and the amount of error accumulated ?

what is the error for pixels where the two two images don't overlap ? any choice you make, will introduce a bias one way or the other.

what am I missing ?
02:12 PM ET (US)
I don't have an idea as to what the author means by:
" We represent the depth map using a tensor product spline".

What I liked about the paper is that they say that its a good idea to bootstrap the dense depth recovery algorithm by first estimating the camera motion using a feature based structure from motion algorithm.

However, I feel that most of the things described in the paper are only standard ways in which these problems can be approached.

Print | RSS Views: 1363 (Unique: 745 ) / Subscribers: 1 | What's this?