From: Ian Scott [mailto:email@example.com]
Sent: Friday, September 20, 2002 10:26 AM
To: Vxl-maintainers (E-mail)
Cc: 'Joseph Mundy'
Subject: RE: [Vxl-maintainers] RE: VXL MeetingCould people reply to the reflector saying if they are coming to the meeting (telephonically or actually).I'd rather not buy a flight unless enough other people actually join in. We are working on some proposal for the modification of vil, and I would like to get some agreement on the changes.Ian.-----Original Message-----
From: Ian Scott
Sent: Thursday, September 19, 2002 6:03 PM
To: 'Joseph Mundy'; Vxl-maintainers (E-mail)
Cc: 'Andrew Fitzgibbon'
Subject: [Vxl-maintainers] RE: VXL MeetingHiI will be coming (assuming I can get a flight etc.) Tim has been flying around too much this month and will join us by phone.Joe, do you know of a suitable hotel that I can book?Tim and I with some help from AWF have been brainstorming a proposal to revamp vil - more details later.Below is a slightly updated list of proposals for agenda items that I put on the reflector some time ago.
Some Agenda Items
* - Indicates ones Manchester are prepared to put time in preparing for / implementing.
Relationship with other librarieshere is ITK going?
Boost - which has random number generators, lambda notation for inline functions, all sorts of smart pointers compile-time concept checking for generic programming. It is likely to be added in part to the STL.
Candidates for promotion to VXL. *
mbl_mz_random -> vnl
mbl_priority_bound_queue -> vbl
mbl_lru_cache -> vbl
mbl_cloneable_ptr -> vbl
mbl_gamma -> vnl
mbl_matxvec -> vnl
mbl_matrix_products -> vnl
mbl_chord -> vil or vgl
mbl_thin_plate_spline -> vgl_algo
mbl_index_sort_cmp -> vbl
mbl_stl -> vbl
pdf1d, vpdfl -> vxl/pdf
Get out a release.
This is important from an "encouraging others to use VXL" point of view.
Several people (including as I recall, Bill Hoffman) have complained about the fact that vcl_vector_fixed is non-optimal. A 2 element vector needs 3 elements of storage (1 for n_dims_)
Options include using MTL, TNT, or upgrading VNL.
We would like to retrofit VIL with some of mil's features, e.g.
Merge concepts of planes and channels. (i.e. assert nplanes == 1 || nchannels == 1)
Using pointer arithmetic based addressing in vil_memory_image.
vil_memory_image_impl should know its type at compile time
Addition of standard polymorphic image and world2image transform type. Could be a specialised vcl_pair.
Possiblity of modifying vil to make its interface easily extendable to 3/4/D (with multiple planes)
Is anyone prepared to write some more.e.g. for vcsl and vbl.
The Dart system is great - but it points out that some of our testing coverage is pretty poor.
The current system is difficult for any but the most competent developers. Perhaps some automatic generating scheme might help.