http:^^www.cs.wisc.edu^~dyer^cs766^hw3-faqs.html
来自「This data set contains WWW-pages collect」· HTML 代码 · 共 118 行
HTML
118 行
Date: Mon, 11 Nov 1996 17:30:50 GMTServer: NCSA/1.5Content-type: text/htmlLast-modified: Tue, 07 Nov 1995 18:56:24 GMTContent-length: 4861<html><head><title>CS 766 HW #3 Hints and FAQs</title></head><body><H1>CS 766 HW #3 Hints and FAQs</H1><P><HR><UL> <LI><B>How is the edge magnitude information in `edge' accessed?</B><BR> The array `edge' should be accessed as edge[row][col]. <P> <LI><B>My snake points never settle down; is that a problem?</B><BR> It can happen, so it's not necessarily a bug. It might mean that there are too many snake points. <P> <LI><B>What happens if two consecutive snake points move to the same pixel?</B><BR> There is a problem with the Williams and Shah algorithm in this case. That is, the curvature at each of those points will be undefined if one follows the algorithm to the letter. To deal with that, you can do the following. When computing the curvature c at point i, check whether the ith point coincides with the previous and/or the next points. If it does, set the curvature of the ith point to BIGC, where BIGC is a large constant. This will ensure that if points i and i+1 are coincident, then for all four points i-1, i, i+1 and i+2, the "if curvature is larger than neighbors" condition of the algorithm will fail and beta won't be changed. (Make sure you use '>' in this condition and not '>='). <P> <LI><B>What if snake point i goes to the old position of point i+1 for all i (modulo the total number of snake points)?</B><BR> You do not have to deal with this problem. If it ever happens when you are developing your code or during the demo, you can simply stop the snake and restart with a new snake. <P> <LI><B>In MoveSnakePoint, does the parameter `new' refer to Points[jmin], and is the parameter `index' the index i into the `points' array, i.e., the point (vector) v_i to be moved to jmin?</B><BR> The parameter 'new' (the first parameter of MoveSnakePoint) is of type Point. It should contain the value of the coordinates of the new position of a point. But if you are moving, say points[i], you should NOT change points[i] directly and then call<pre> MoveSnakePoint(points[i], points, i, num_points)</pre> Rather, you should create a temporary Point variable, say loc_min, define loc_min.row and loc_min.col appropriately and then call<pre> MoveSnakePoint(loc_min, points, i, num_points)</pre> MoveSnakePoint() is itself supposed to copy the coordinates from loc_min to points[i] in addition to moving the point on the display. <P> <LI><B>When calculating the derivatives for Econt and Ecurv, how should the boundary conditions, i.e., when i=0 and i=n, be handled?</B><BR> For this assignment you can assume all contours are closed, and therefore you can use modulo n+1 arithmetic when dealing with snaxel indices. Hence, v(0) is adjacent to v(n). <P> <LI><B>The snake program interface displays poor colors, what's wrong?</B><BR> Your window manager may be setting default color settings differently than needed by the snakes program. Install a different .fvwmrc file, for example, to change these settings. <P> <LI><B>How is Econt calculated?</B><BR> On page 19 in the Williams and Shah paper there is a description of how to compute this term. There is an error, however. Since we want this to be a measure of distance, you should use: <pre> - | d - | v - v | | i i-1 </pre> To normalize this value to be in the range [0..1], divide by the largest value of this expression from the 9 values computed for the 9 possible positions of v_i in the 3 x 3 neighborhood centered at the current point. <P> <LI><B>For both the continuity and curvature energies, the paper states thatthe value at a point under consideration should be divided by the largest valuein the neighborhood to which the point may move. I am interpreting this tomean the largest value in the neighborhood around the original snake point. Someone else interpreted this to mean you figure out the largest value in the neighborhood for the point under consideration. For example:<pre>+-+-+-+|1|2|3|+-+-+-+-+ X = the snake point|4|*|5| | * = the point under consideration+-+-+-+-+|6|7|X| | So, the question is: which largest value do I use? The one that+-+-+-+-+ is computed using the neighborhood around X or do I compute based | | | | upon the neighborhood of the point under consideration? So, for +-+-+-+ example, if I wanted to normalize the value computed at *, I would figure out the largest value at 1,2,3,4,*,5,6,7,X becuase that is *'s neighborhood.</pre>Which is correct?</B><BR>The way I read the paper * should be normalized by using the largest valuefrom the 9 values in the 3 x 3 neighborhood centered at X, not *. Otherwise,each point would be normalized differently, making comparison difficult.<P></UL></body></html>⌨️ 快捷键说明
复制代码Ctrl + C
搜索代码Ctrl + F
全屏模式F11
增大字号Ctrl + =
减小字号Ctrl + -
显示快捷键?