def make_mask(self, threshisl=5):
"""
Create a mask of the image where only belivable flux is
"""
logger.info('%s: Making mask...' % self.imagename)
if not os.path.exists(self.maskname):
make_mask(image_name=self.imagename, mask_name=self.maskname, threshisl=threshisl, atrous_do=True)
if self.user_mask is not None:
logger.info('%s: Adding user mask (%s)...' % (self.imagename, self.user_mask))
blank_image_reg(self.maskname, self.user_mask, inverse=False, blankval=1)
def ft_model_cc(mss,
imagename,
c,
user_mask=None,
keep_in_beam=True,
model_column='MODEL_DATA'):
"""
skymodel : cc-list made by wsclean
keep_in_beam : if True remove everything outside primary beam, otherwise everything inside
"""
logger.info('Predict with CC...')
maskname = imagename + '-mask.fits'
skymodel = imagename + '-sources.txt'
skymodel_cut = imagename + '-sources-cut.txt'
skydb = imagename + '-sources.skydb'
# prepare mask
if not os.path.exists(maskname):
logger.info('Predict (make mask)...')
make_mask(image_name=imagename + '-MFS-image.fits',
mask_name=maskname,
threshisl=5,
atrous_do=True)
if user_mask is not None:
blank_image_reg(maskname, user_mask, inverse=False,
blankval=1) # set to 1 pixels into user_mask
blank_image_reg(
maskname, 'self/beam.reg', inverse=keep_in_beam,
blankval=0) # if keep_in_beam set to 0 everything outside beam.reg
# apply mask
logger.info('Predict (apply mask)...')
lsm = lsmtool.load(skymodel)
lsm.select('%s == True' % maskname)
fluxes = lsm.getColValues('I')
#lsm.remove(np.abs(fluxes) < 5e-4) # TEST
lsm.write(skymodel_cut, format='makesourcedb', clobber=True)
del lsm
# convert to skydb
logger.info('Predict (makesourcedb)...')
check_rm(skydb)
s.add('makesourcedb outtype="blob" format="<" in="' + skymodel_cut +
'" out="' + skydb + '"',
log='makesourcedb-c' + str(c) + '.log',
cmd_type='general')
s.run(check=True)
# predict
logger.info('Predict (ft)...')
for ms in mss:
s.add('NDPPP '+parset_dir+'/NDPPP-predict.parset msin='+ms+' msout.datacolumn='+model_column+' pre.usebeammodel=false pre.sourcedb='+skydb, \
log=ms+'_pre-c'+str(c)+'.log', cmd_type='NDPPP')
s.run(check=True)
def ft_model_wsclean(mss,
imagename,
c,
user_mask=None,
keep_in_beam=True,
resamp=None,
model_column='MODEL_DATA'):
"""
mss : vector of mss
imagename : root name for wsclean model images
resamp : must be '10asec' or another pixels size to resample models
keep_in_beam : if True remove everything outside primary beam, otherwise everything inside
"""
logger.info('Predict with model image...')
# remove CC not in mask
logger.info('Predict (mask)...')
maskname = imagename + '-mask.fits'
make_mask(image_name=imagename + '-MFS-image.fits',
mask_name=maskname,
threshisl=5,
atrous_do=True)
if user_mask is not None:
blank_image_reg(maskname, user_mask, inverse=False, blankval=1)
blank_image_reg(maskname, 'self/beam.reg', inverse=keep_in_beam)
for modelname in sorted(glob.glob(imagename + '*model.fits')):
blank_image_fits(modelname, maskname, inverse=True)
if resamp is not None:
logger.info('Predict (resamp)...')
for model in sorted(glob.glob(imagename + '*model.fits')):
model_out = model.replace(imagename, imagename + '-resamp')
s.add('/home/fdg/opt/src/nnradd/build/nnradd ' + resamp + ' ' +
model_out + ' ' + model,
log='resamp-c' + str(c) + '.log',
log_append=True,
cmd_type='general')
s.run(check=True)
imagename = imagename + '-resamp'
logger.info('Predict (ft)...')
s.add('wsclean -predict -name ' + imagename + ' -mem 90 -j '+str(s.max_processors)+' -channelsout 10 '+' '.join(mss), \
log='wscleanPRE-c'+str(c)+'.log', cmd_type='wsclean', processors='max')
s.run(check=True)
if model_column != 'MODEL_DATA':
logger.info('Predict (set %s = MODEL_DATA)...' % model_column)
for ms in mss:
s.add('taql "update ' + ms + ' set ' + model_column +
' = MODEL_DATA"',
log=ms + '_taql0-c' + str(c) + '.log',
cmd_type='general')
s.run(check=True)
def inference(self, question, image_feat):
"""Function used for running inverence given the image and question
Args:
question (str): The question
image (np.ndarray): The image
Returns:
[dict]: {
"img": {
"iatt_maps": torch.Tensor(1, 100, 2)
},
"text" {
"qatt": torch.Tensor(1, 14, 2)
}
}
"""
# extract indices of question words
ques_ix = proc_ques(question, self.token_to_ix, max_token=14)
ques_ix = torch.tensor(ques_ix).unsqueeze(0)
# pad image features
frcn_feat = torch.zeros(100, 2048)
frcn_feat[:image_feat.shape[0]] = image_feat
image_feat = frcn_feat.unsqueeze(0)
image_feat_mask = make_mask(image_feat)
ret = self.net.forward(image_feat, image_feat_mask, ques_ix)
return ret
def forward(self, img_feat, img_feat_mask, ques_ix, **kwargs):
# Pre-process Language Feature
ques_feat_mask = make_mask(ques_ix.unsqueeze(2)).squeeze(2).permute(
0, 2, 1)
ques_feat = self.embedding(ques_ix) # (N, T, WORD_EMBED_SIZE)
ques_feat = self.dropout(ques_feat)
ques_feat, _ = self.lstm(ques_feat) # (N, T, LSTM_OUT_SIZE)
ques_feat = self.dropout_lstm(ques_feat)
text_ret = defaultdict(list)
img_ret = defaultdict(list)
z = self.backbone(img_feat,
ques_feat,
img_feat_mask.squeeze(2).permute(0, 2, 1),
ques_feat_mask,
text_ret=text_ret,
img_ret=img_ret) # MFH:(N, 2*O) / MFB:(N, O)
proj_feat = self.proj(z) # (N, answer_size)
return_dict = {
"proj_feat": proj_feat,
"img": img_ret,
"text": text_ret
}
return return_dict
###########################################################################################################################
# avg time only - SB.MS:CORRECTED_DATA -> concat-avg.MS:DATA (selfcal corrected data, beam applied, circular)
logging.info('Average...')
check_rm('concat-avg.MS*')
s.add('NDPPP /home/fdg/scripts/autocal/VirgoA/parset_self-lowres/NDPPP-concatavg2.parset msin="['+','.join(mss)+']" msout=concat-avg.MS avg.freqstep=1 avg.timestep=4', \
log='concatavg.log', cmd_type='NDPPP')
s.run(check=True)
#########################################################################################################
# large FoV image
logging.info('Low-res wide field image...')
s.add_casa('/home/fdg/scripts/autocal/casa_comm/virgoLBA/casa_clean-lowres.py', \
params={'msfile':'concat-avg.MS', 'imagename':'img/clean-wide', 'imtype':'wide'}, log='clean-wide.log')
s.run(check=True)
make_mask(image_name='img/clean-wide.image.tt0', mask_name='img/wide.newmask')
logging.info('Low-res wide field image (with mask)...')
s.add_casa('/home/fdg/scripts/autocal/casa_comm/virgoLBA/casa_clean-lowres.py', \
params={'msfile':'concat-avg.MS', 'imagename':'img/clean-wide-masked', 'imtype':'wide', 'mask':'img/wide.newmask'}, log='clean-wide-masked.log')
#########################################################################################################
# central clean
logging.info('Clean...')
s.add_casa('/home/fdg/scripts/autocal/casa_comm/virgoLBA/casa_clean-lowres.py', \
params={'msfile':'concat-avg.MS', 'imagename':'img/normal'}, log='clean.log')
s.run(check=True)
##########################################################################################################
# uvsub + large FoV image
s.add(
'taql "update concat-avg.MS set CORRECTED_DATA = CORRECTED_DATA - MODEL_DATA"'
def clean(c, mss, dd, avgfreq=4, avgtime=10, facet=False, skip_mask=False):
"""
c = cycle/name
mss = list of mss to avg/clean
"""
# averaging before cleaning *.MS:CORRECTED_DATA -> *-avg.MS:DATA
logging.info('Averaging before cleaning...')
check_rm('mss_imgavg')
os.mkdir('mss_imgavg')
nchan = find_nchan(mss[0])
for ms in mss:
msout = 'mss_imgavg/'+os.path.basename(ms)
check_rm(msout)
s.add('NDPPP '+parset_dir+'/NDPPP-avg.parset msin='+ms+' msin.nchan='+str(nchan-nchan%4)+' msin.datacolumn=CORRECTED_DATA \
msout='+msout+' avg.freqstep='+str(avgfreq)+' avg.timestep='+str(avgtime), log=ms+'_cleanavg-c'+str(c)+'.log', cmd_type='NDPPP')
s.run(check=True)
mssavg = [ms for ms in glob.glob('mss_imgavg/*MS')]
# set image name
imagename = 'peel/'+dd['name']+'/images/peel-'+str(c)
# set pixscale, imsize and niter
pixscale = scale_from_ms(mss[0])
if facet:
imsize = size_from_reg('peel/'+dd['name']+'/models/peel_facet-0000-model.fits', 'regions/'+dd['reg_facet'], dd['coord'], pixscale, pad=1.5)
# niter = 10000
else:
imsize = size_from_reg('peel/'+dd['name']+'/models/peel_dd-0000-model.fits', 'regions/'+dd['reg'], dd['coord'], pixscale, pad=1.5)
# niter = 2000
if imsize < 512:
trim = 512
imsize = 1024
elif imsize < 1024:
trim = 1024
imsize = 1024
else:
trim = imsize
logging.debug('Image size: '+str(imsize)+' - Pixel scale: '+str(pixscale))
# TODO: add multiscale
if dd['extended']: multiscale = [0,4,16,64]
else: multiscale = []
# Clean mask clean
logging.info('Cleaning (cycle: '+str(c)+')...')
s.add('wsclean -reorder -name ' + imagename + ' -size '+str(imsize)+' '+str(imsize)+' -trim '+str(trim)+' '+str(trim)+' \
-mem 90 -j '+str(s.max_processors)+' -baseline-averaging 2.0 \
-scale '+str(pixscale)+'arcsec -weight briggs 0.0 -threshold 0.05 -niter 10000 -no-update-model-required -mgain 0.6 \
-pol I -cleanborder 0 -joinchannels -fit-spectral-pol 2 -channelsout 10 -deconvolution-channels 5 '+' '.join(mss), \
log='wsclean-c'+str(c)+'.log', cmd_type='wsclean', processors='max')
s.run(check=True)
if skip_mask:
check_rm('mss_imgavg')
return imagename, trim, pixscale
if dd['faint'] or facet:
make_mask(image_name = imagename+'-MFS-image.fits', mask_name = imagename+'.newmask', atrous_do=dd['extended'], threshisl=5)
else:
make_mask(image_name = imagename+'-MFS-image.fits', mask_name = imagename+'.newmask', atrous_do=dd['extended'], threshisl=20)
# if dd['mask'] is set then add it to the new mask
if dd['mask'] != '':
s.add_casa('/home/fdg/scripts/autocal/casa_comm/casa_blank.py', \
params={'imgs':imagename+'.newmask', 'region':dd['mask'], 'setTo':1}, log='casablank-c'+str(c)+'.log', log_append=True)
s.run(check=True)
logging.info('Cleaning with mask (cycle: '+str(c)+')...')
s.add('wsclean -reorder -name ' + imagename + '-M -size '+str(imsize)+' '+str(imsize)+' -trim '+str(trim)+' '+str(trim)+' \
-mem 90 -j '+str(s.max_processors)+' -baseline-averaging 2.0 -casamask '+imagename+'.newmask \
-scale '+str(pixscale)+'arcsec -weight briggs 0.0 -threshold 0.05 -niter 5000 -no-update-model-required -mgain 0.6 \
-pol I -cleanborder 0 -joinchannels -fit-spectral-pol 2 -channelsout 10 -deconvolution-channels 5 '+' '.join(mss), \
log='wscleanM-c'+str(c)+'.log', cmd_type='wsclean', processors='max')
s.run(check=True)
check_rm('mss_imgavg')
return imagename, trim, pixscale
###########################################################################################################################
# avg time only - SB.MS:CORRECTED_DATA -> concat-avg.MS:DATA (selfcal corrected data, beam applied, circular)
logging.info('Average...')
check_rm('concat-avg.MS*')
s.add('NDPPP /home/fdg/scripts/autocal/VirgoA/parset_self-lowres/NDPPP-concatavg2.parset msin="['+','.join(mss)+']" msout=concat-avg.MS avg.freqstep=1 avg.timestep=4', \
log='concatavg.log', cmd_type='NDPPP')
s.run(check=True)
#########################################################################################################
# large FoV image
logging.info('Low-res wide field image...')
s.add_casa('/home/fdg/scripts/autocal/casa_comm/virgoLBA/casa_clean-lowres.py', \
params={'msfile':'concat-avg.MS', 'imagename':'img/clean-wide', 'imtype':'wide'}, log='clean-wide.log')
s.run(check=True)
make_mask(image_name = 'img/clean-wide.image.tt0', mask_name = 'img/wide.newmask')
logging.info('Low-res wide field image (with mask)...')
s.add_casa('/home/fdg/scripts/autocal/casa_comm/virgoLBA/casa_clean-lowres.py', \
params={'msfile':'concat-avg.MS', 'imagename':'img/clean-wide-masked', 'imtype':'wide', 'mask':'img/wide.newmask'}, log='clean-wide-masked.log')
#########################################################################################################
# central clean
logging.info('Clean...')
s.add_casa('/home/fdg/scripts/autocal/casa_comm/virgoLBA/casa_clean-lowres.py', \
params={'msfile':'concat-avg.MS', 'imagename':'img/normal'}, log='clean.log')
s.run(check=True)
##########################################################################################################
# uvsub + large FoV image
s.add('taql "update concat-avg.MS set CORRECTED_DATA = CORRECTED_DATA - MODEL_DATA"') # uvsub
s.run(check=False)
s.add('taql "update '+ms+' set CORRECTED_DATA = CORRECTED_DATA - MODEL_DATA"', log=ms+'_taql2.log', cmd_type='general')
s.run(check=True)
logger.info('Cleaning...')
check_rm('img')
os.makedirs('img')
imagename = 'img/wide'
s.add('wsclean -reorder -name ' + imagename + ' -size 5000 5000 -trim 4000 4000 -mem 90 -j '+str(s.max_processors)+' -baseline-averaging 2.0 \
-scale 5arcsec -weight briggs 0.0 -niter 100000 -no-update-model-required -mgain 0.9 \
-pol I -joinchannels -fit-spectral-pol 2 -channelsout 10 -auto-threshold 20 '+' '.join(mss), \
log='wscleanA.log', cmd_type='wsclean', processors='max')
s.run(check=True)
# make mask
maskname = imagename+'-mask.fits'
make_mask(image_name = imagename+'-MFS-image.fits', mask_name = maskname, threshisl = 3, atrous_do=True)
# remove CC not in mask
for modelname in sorted(glob.glob(imagename+'*model.fits')):
blank_image_fits(modelname, maskname, inverse=True)
logger.info('Cleaning w/ mask')
imagename = 'img/wideM'
s.add('wsclean -reorder -name ' + imagename + ' -size 5000 5000 -trim 4000 4000 -mem 90 -j '+str(s.max_processors)+' -baseline-averaging 2.0 \
-scale 5arcsec -weight briggs 0.0 -niter 100000 -no-update-model-required -mgain 0.8 -minuv-l 30 \
-pol I -joinchannels -fit-spectral-pol 2 -channelsout 10 -auto-threshold 0.1 -save-source-list -apply-primary-beam -use-differential-lofar-beam \
-fitsmask '+maskname+' '+' '.join(mss), \
log='wscleanB.log', cmd_type='wsclean', processors='max')
s.run(check=True)
# prepare mask
logger.info('Masking skymodel...')
def clean(c, mss, dd, avgfreq=2, avgtime=8, facet=False):
"""
c = cycle/name
mss = list of mss to avg/clean
"""
# averaging before cleaning *.MS:CORRECTED_DATA -> *-avg.MS:DATA
logger.info('Averaging before cleaning...')
check_rm('mss_imgavg')
os.mkdir('mss_imgavg')
nchan = find_nchan(mss[0])
for ms in mss:
msout = 'mss_imgavg/' + os.path.basename(ms)
check_rm(msout)
s.add('NDPPP ' + parset_dir + '/NDPPP-avg.parset msin=' + ms +
' msin.nchan=' + str(nchan - nchan % 4) +
' msin.datacolumn=CORRECTED_DATA \
msout=' + msout + ' avg.freqstep=' + str(avgfreq) +
' avg.timestep=' + str(avgtime),
log=ms + '_cleanavg-c' + str(c) + '.log',
cmd_type='NDPPP')
s.run(check=True)
mssavg = [ms for ms in sorted(glob.glob('mss_imgavg/*MS'))]
# set pixscale and imsize
pixscale = scale_from_ms(mssavg[0])
if facet:
imsize = int((dd['facet_size'] / (pixscale / 3600.)) * 1.5)
else:
imsize = int((dd['dd_size'] / (pixscale / 3600.)) * 1.5)
if imsize < 512:
imsize = 512
trim = int(imsize * 0.7)
if imsize % 2 == 1: imsize += 1 # make even
if trim % 2 == 1: trim += 1 # make even
logger.debug('Image size: ' + str(imsize) + ' - Pixel scale: ' +
str(pixscale))
# -trim '+str(trim)+' '+str(trim)+'
# -auto-mask 5 -auto-threshold 1 -rms-background -rms-background-window 25 \
# -multiscale
# clean 1
logger.info('Cleaning (cycle: ' + str(c) + ')...')
if facet: imagename = 'img/facet-' + str(c)
else: imagename = 'img/ddcal-' + str(c)
s.add('/home/fdg/opt/src/wsclean-2.2.9/build/wsclean -reorder -name ' + imagename + ' -size '+str(imsize)+' '+str(imsize)+' -trim '+str(trim)+' '+str(trim)+' \
-mem 90 -j ' +str(s.max_processors)+' -baseline-averaging 2.0 \
-scale ' +str(pixscale)+'arcsec -weight briggs 0.0 -niter 100000 -no-update-model-required -mgain 0.9 -pol I \
-joinchannels -fit-spectral-pol 2 -channelsout 10 \
-auto-threshold 20 ' +' '.join(mssavg), \
log='wsclean-c'+str(c)+'.log', cmd_type='wsclean', processors='max')
s.run(check=True)
os.system('cat logs/wsclean-c' + str(c) + '.log | grep Jy')
# make mask
maskname = imagename + '-mask.fits'
make_mask(image_name=imagename + '-MFS-image.fits',
mask_name=maskname,
threshisl=3)
# clean 2
logger.info('Cleaning w/ mask (cycle: ' + str(c) + ')...')
if facet: imagename = 'img/facetM-' + str(c)
else: imagename = 'img/ddcalM-' + str(c)
s.add('/home/fdg/opt/src/wsclean-2.2.9/build/wsclean -reorder -name ' + imagename + ' -size '+str(imsize)+' '+str(imsize)+' -trim '+str(trim)+' '+str(trim)+' \
-mem 90 -j ' +str(s.max_processors)+' -baseline-averaging 2.0 \
-scale ' +str(pixscale)+'arcsec -weight briggs 0.0 -niter 100000 -no-update-model-required -mgain 0.8 -pol I \
-joinchannels -fit-spectral-pol 2 -channelsout 10 \
-auto-threshold 0.1 -fitsmask ' +maskname+' '+' '.join(mssavg), \
log='wscleanM-c'+str(c)+'.log', cmd_type='wsclean', processors='max')
s.run(check=True)
os.system('cat logs/wscleanM-c' + str(c) + '.log | grep Jy')
# remove CC not in mask
maskname = imagename + '-mask.fits'
if facet:
make_mask(image_name=imagename + '-MFS-image.fits',
mask_name=maskname,
threshisl=5)
else:
make_mask(image_name=imagename + '-MFS-image.fits',
mask_name=maskname,
threshisl=7)
for modelname in sorted(glob.glob(imagename + '*model.fits')):
blank_image_fits(modelname, maskname, inverse=True)
check_rm('mss_imgavg')
return imagename
# concatenate data - MS.MS -> concat.MS (selfcal corrected data, beam applied, circular)
logging.info('Make widefield model - Concatenating...')
check_rm('concat.MS*')
pt.msutil.msconcat(mss_c, 'concat.MS', concatTime=False)
# uvsub, MODEL_DATA is still Virgo
logging.info('Make widefield model - UV-Subtracting Virgo A...')
os.system('taql "update concat.MS set CORRECTED_DATA = CORRECTED_DATA - MODEL_DATA"') # uvsub
# clean, mask, clean
logging.info('Make widefield model - Widefield imaging...')
imagename = 'img/clean-wide-c'+str(i)
s.add_casa('/home/stsf309/scripts/autocal/casa_comm/virgoHBA/casa_clean.py', \
params={'msfile':'concat.MS', imagename=imagename, imtype='wide'}, log='clean-wide1-c'+str(i)+'.log')
s.run(check=True)
make_mask(image_name = imagename+'.image.tt0', mask_name = imagename+'.newmask')
s.add_casa('/home/stsf309/scripts/autocal/casa_comm/casa_blank.py', params={'imgs':imagename+'.newmask', 'region':'/home/stsf309/scripts/autocal/VirA_HBA/m87.crtf'})
s.run(check=True)
logging.info('Make widefield model - Widefield imaging2...')
s.add_casa('/home/stsf309/scripts/autocal/casa_comm/virgoHBA/casa_clean.py', \
params={'msfile':'concat.MS', imagename=imagename.reaplce('wide','wode-masked'), mask=imagename+'.newmask' imtype='wide'}, log='clean-wide2-c'+str(i)+'.log')
s.run(check=True)
# Subtract widefield model using ft on a virtual concat - concat.MS:CORRECTED_DATA -> concat.MS:CORRECTED_DATA-MODEL_DATA (selfcal corrected data, beam applied, circular, field sources subtracted)
logging.info('Flagging - Subtracting wide-field model...')
check_rm('concat.MS*')
pt.msutil.msconcat(mss_c, 'concat.MS', concatTime=False)
s.add_casa('/home/stsf309/scripts/autocal/casa_comm/casa_ft.py', \
params={'msfile':'concat.MS', 'model':'img/wide-c'+str(i)+'.model', 'wproj':512}, log='ft-flag-c'+str(i)+'.log')
s.run(check=True)
os.system('taql "update concat.MS set CORRECTED_DATA = CORRECTED_DATA - MODEL_DATA"') # uvsub
