home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: sci.image.processing
- Path: sparky!uunet!cs.utexas.edu!zaphod.mps.ohio-state.edu!news.acns.nwu.edu!casbah.acns.nwu.edu!phantom
- From: phantom@casbah.acns.nwu.edu (James Choi)
- Subject: Re: Debugging problem
- Message-ID: <1993Jan26.014330.29841@news.acns.nwu.edu>
- Sender: usenet@news.acns.nwu.edu (Usenet on news.acns)
- Nntp-Posting-Host: unseen1.acns.nwu.edu
- Organization: Northwestern University, Evanston Illinois.
- References: <1993Jan26.012844.15569@usl.edu>
- Date: Tue, 26 Jan 1993 01:43:30 GMT
- Lines: 32
-
- In article <1993Jan26.012844.15569@usl.edu> twc@swamp.cacs.usl.edu (Tat W. Chan) writes:
- > 1. Debugging a large program which input data is huge and not
- > well-structured (as it is common in the domain of this newsgroup),
- > is there any effective ways to locate the bugs? Should we deal with
- > the program first, or the data first, or both?
- >
- > 2. If there is any good way to solve #1, is it different from debugging
- > techniques in other domains in which the input data is not so intensive?
-
- When debugging a typical image processing algorithm, conventional debuggers
- are of little help. The vast amount of data that image generates makes it
- fruitless to examine each number.
-
- The most useful way I know of in dealing with is using a model data before
- you use a real data set. For example, if you are looking for a centroid,
- use a picture that consist of 0 and 255 only, and that forms a perfect
- square. Of course, this data is generated, not acquired. If the square
- works, then try a rectangle, and so on.
-
- In CT image reconstruction algorithm testing, I use either a thin wire to
- see if the algorithm reconstructs a point. Or better yet, use simulated
- data.
-
- Some algorithms take hours, or even days in development stage. Use lower
- resolution in debugging stage.
-
- Have a standard set of simulated data that you use every time you make any
- modification to the algorithm.
-
- That's all I can remember at the moment.
-
- James Choi
-