floret DAISY parallel multi-threaded object classifier
Although it is possible to use
floret(1)
in a standalone mode, it is designed to be used in conjunction with the
DAISY
frontend,
dfe(1).
This is a
Gtk
based graphical frontend, which allows the user to capture images (which are to be classified
by
floret(1)
) from a variety of sources including microscopes, document scanners and gel electrophoresis
imaging systems.
dfe(1)
also allows the user to manipulate the captured image data before it is passed to
floret(1).
Typically this manipulation consists of radiometric normalisation, standard pose recovery
and scaling.
-growth_protection
<retrys>
causes the system to retry an
sbrk(3)
call up to
retrys
times. This option permits the program to wait until the core that it requires
is available rather than simply exiting if
sbrk(3)
fails.
-usage
displays shortform usage string.
-version
displays program version number.
-slots
displays short form
pups(7)
library dependencies.
-slotinfo
displays long form
pups(3)
library dependencies.
-argf
argument_file
tells the filter to take its command line arguments from the specified
argument_file.
The data format of the argument file is:
# ... optional comment line ...
-argflag [1]
argument_parameter [1,1]
argument_parameter [1,2]
argument_parameter [1,n]
# ... optional comment line ...
-argflag [2]
argument_parameter [1,1]
argument_parameter [1,2]
argument_parameter [1,n]
-nice
cardinal
determimes the
niceness
at which the filter is scheduled. In keeping with all UNIX user processes, nice
must be a cardinal between 0 and 20. The default niceness for
floret(1)
is 4.
-noprogress
tells
ipm(1)
not to output information for the
progress(1)
-migrate
pafname
tells
floret(1)
to migrate to the host whose file system contains its training sets. This improves
performance as access to training sets is not over slow network filesystem protocols.
If
pafname
is supplied
ssh(1)
will authenticate the migration from this
PSRP authenitcation FIFO (PAF)
if not, the information will be acquired interactively via
floret(1)'s
controlling tty.
-tdir training set directory is used to set the directory in which training images are located. This option may also be changed interatively via a PSRP session (dispatch binding is also tdir ). This option is mandatory and must be supplied.
-sdir
sensitive directory list
is used to set the list of directories in which
dfe(1)
may place images to be classified. Te directory list may be modified
via a
PSRP
session (dispatch binding is also
sdir
). This option is mandatory and must be supplied.
-tpool training pool directory tells floret(1) that a training pool exists (and is to be homeostatically proctected by this invocation of floret(1) ). br
-prune
maximum training set size
tells
floret(1)
to prune the training set. If more than
maximum training set size
objects are present in the training set random objects are selected and deleted
until the training set size is less than
maximum training set size
-key
floret t.s. I.D. key
tells
floret(1)
to only compare unknown images against training set images which match
floret t.s. I.D. key.
This option is provided to facilitate parallel
DAISY
implementations where each
infloresence class
is responsible for the classification of part of the dataset.
The key may also be set via an interactive PSRP session
(binding for PSRP dispatch function is also
key
).
-identify
unknown image name
tells
floret(1)
to identify the unknown image immediately and then exit. A
PUPS/PSRP
dispatch function (with identical name) also exists. In this case
floret(1)
interupts its monitoring of its sensitive directory to immediately classify
the image presented to it via the
PSRP
API. This flag and the
-sdir
flag are mutually exclusive.
-no_output
tells
floret(1)
not to produce any output when doing an immediate classification. This option is
used when
floret(1)
is slaved by
jack(1)
or similar applications. This option is only valid if
-identify
has been specified.
-no_output
and
-mlst
are mutually exclusive.
-mlst
matchlist file pathname
tells
floret(1)
to use
matchlist file pathname
for the matchlist filename path. Typically this is used by applications such as
jack(1)
the training set jacknife (identification) statisitics generator, which need unique
matchlist pathnames. This option is only valid if
-identify
has been specified.
-mlst
and
-no_output
are mutually exclusive.
-class
tells
floret(1)
to build a matchlist of the
matchlist size
nearest classes (as opposed to the
matchlist size
nearest
objects.
Defaults to FALSE (matchlist of objects is generated).
-mlsize
matchlist size
sets the size of the
matchlist
produced by this invocation of
floret(1).
Defaults to 8.
-lifetime
seconds
tells
floret(1)
to retain classified unknowns in its sensitive directory for
seconds.
This option is provided so that
vhtml(1)
data output functions have sufficient time to inspect the classified
image and (if possible) generate a data page for it. This option may
also be set via an interactive PSRP session (dipatch function binding is
lifetime
).
-npf
florets
tells
floret(1)
the number of florets which are co-operating with it to identify the
ciurrent unknown. A group of co-operating florets is known as an
infloresence.
This option may
also be set via an interactive PSRP session (dipatch function binding is
npf
). Defaults to one.
-rgb
RGB
tells
floret(1)
which bands of an RGB image it is to take into account when classifying an
unknown. This option may
also be set via an interactive PSRP session (dipatch function binding is
mask
).
-in_state
computational input state tag
tells
floret(1)
to set its input computational state tage to
computational input state tag. Defaults to
floret.in
-out_state
computational output state tag
tells
floret(1)
to set its input computational state tag to
computational output state tag.
Defaults to
floret.out
-psom
learning rate
tells
floret(1)
to use PSOM (plastic self organising map) dynamic learning. If no learning rate is specified
the default learning rate (0.01) is used.
-mask
tells
floret(1)
that manifold reduction masks are available. These masks are generated via an annealing
process. The masks are used in order to more effectively cluster training set images
within morph space.
-max_offset
offset size
sets the offset
fuzz factor
for unposed images. The correlations of the unknown image and best training set image will
be computed with the first azimuthal bin of the unknown image at
-fuzz_offset
to
offset
inclusive. The
true
correlation is then assumed to be the correlation (of the 2*offset + 1 possible correlations) which
exhibits the highest NVD affinity. Defaults to
posed.
-rgen
rgen_path_name
tells
floret(1)
to append rgenlist data to
rgen_path_name.
Date is subsequently abstracted from the file by
rgen(1)
and passed to
smboltzmann(1)
for automated clustering to taxanomic units.
-new
tells
floret(1)
to create a new rgen input file. This flag is only valid if
-rgen
is also set.
-head
tells
floret(1)
to write out header information to the rgen input file. This flag is only valid if
-rgen
is also set.
The following functions are available via the PSRP API but not via the command line
interface:
deactivate
tells
floret(1)
to stop processing (unknown) images.
reactivate tells floret(1) to restart processing of (unknown) images. br
These functions are provided to temporarily stop processing while housekeeping tasks
(e.g. binding of dynamic correlator or normalisor functions) is performed via the
PSRP interface.
Note environmental variables are ignored if the corresponding command line option
is provided.
FLORET_NOPROGRESS
do not generate status information for
progress(1)
FLORET_MAX_TDEPTH
depth
sets the
depth
for depth first seach. This option is only valid for the
hierachical
version of floret.
FLORET_MIGRATE
pafname
tells
floret(1)
to migrate to be near filesystem resources using
pafname
to authenticate migration.
FLORET_SDIR:
sensitive directory list
set up
floret(1)
sensetive directory list.
FLORET_TDIR:
training set directory name
set
floret(1)
training set directory.
FLORET_TPOOL
training pool directory name
set
I floret(1)
training pool directory.
FLORET_PRUNE
maximum training set size
telss
floret(1)
to (randomly) prune the training set so that its size is less than
maximum training set size
FLORET_DLL
set
floret(1)
DLL
(from which dynamic correlator/normalisor functions are imported).
FLORET_CFUNC:
correlation function
set
floret(1)
correlation function
(imported from
DLL
(PUPS Orifices format).
).
FLORET_NFUNC: normalisation function set floret(1) normalisation function (imported from DLL (PUPS Orifices format). ).
FLORET_RGB_MASK
RGB
set
floret(1)
colour planes.
FLORET_NPF
n_florets
set number of
florets
in
inflorescence.
FLORET_KEY:
key
set
floret(1)
training set
key.
FLORET_MLSIZE
match list size
sets the upper bound for matchlist for this instance of
floret(1).
FLORET_INSTATE
set computational input state tag.
FLORET_OUTSTATE
set computational output state tag.
FLORET_MAX_LIST_SIZE
sets upper bound of matchlist
FLORET_LIFETIME
sets lifetime of matchlist (in seconds).
FLORET_CMODE
set correlation mode (individual images or class).
FLORET_RGB_MASK
set RGB mask (RGB correlation modes only).
FLORET_SILL
enables dynamic sill optimisation.
FLORET_MEXTHRESH
enables dynamic excitation threshold optimisation.
FLORET_CLASS
enable class (as opposed to individual object based) clustering.
FLORET_PSOM
enable PSOM learning (and optionally set learning rate).
FLORET_MAX_OFFSET
offset
set the fuzz offset (for non posed imagery).
FLORET_RGEN_PATHNAME
rgen_path_name
tells
floret(1)
to append matchlist to
rgen_path_name
for subsequent automatic clustering (to taxanomic units).
FLORET_RGEN_NEW
tells
floret(1)
to create a new rgen matchlist file. This environment variable is only valid
if
FLORET_RGEN_PATHNAME is also set.
FLORET_RGEN_HEAD
tells
floret(1)
to write header rgen matchlist file. This environment variable is only valid
if
FLORET_RGEN_PATHNAME is also set.
rgb_mask
set RGB mask bits for colour image processing.
status
report status of current invocation of
floret(1)
deactivate
deactivate this invocation of
floret(1)
reactivate
reactivate this invocation of
floret(1)
key
set training set image keylist (only training images which match keylist will be considered in classification).
instate set computational input state tag. br
outstate
set computational output state tag
sdir
set sensetive directory name.
tdir
set training set directory name.
identify
immediate identification of unknown supplied via PSRP API.
npf
set number of concurrent florets in DAISY cluster
mlsize
set size of matchlist (upper bound).
lifetime
set lifetime of matchlist (in seconds).
mask
enable dynamic training set image masking (manifold reduction).
psom
enable PSOM learning.
cmode
set clustering mode (individual objects or classes).
max_offset
set offset for fuzzed matching of non-posed imagery.