From b3ec68b57936407e1de7fd3114992c68db0d4817 Mon Sep 17 00:00:00 2001
From: "Rob Knop (Nersc)" <raknop@lbl.gov>
Date: Tue, 22 Oct 2024 09:49:04 -0700
Subject: [PATCH 1/3] Workaround in PYSEx.py to deal with race condition
writing and reading fits; debug logging.
---
sfft/utils/SExSkySubtract.py | 26 ++-
sfft/utils/pyAstroMatic/PYSEx.py | 340 ++++++++++++++++++-------------
2 files changed, 223 insertions(+), 143 deletions(-)
diff --git a/sfft/utils/SExSkySubtract.py b/sfft/utils/SExSkySubtract.py
index d1d4939..d3530b2 100644
--- a/sfft/utils/SExSkySubtract.py
+++ b/sfft/utils/SExSkySubtract.py
@@ -6,6 +6,17 @@
from sfft.utils.pyAstroMatic.PYSEx import PY_SEx
# version: Apr 22, 2024
+import sys
+import logging
+import multiprocessing
+_logger = logging.getLogger(f'sfft')
+if not _logger.hasHandlers():
+ log_out = logging.StreamHandler(sys.stderr)
+ formatter = logging.Formatter(f'[%(asctime)s - sfft - %(levelname)s] %(message)s', datefmt='%Y-%m-%d %H:%M:%S')
+ log_out.setFormatter(formatter)
+ _logger.addHandler(log_out)
+ _logger.setLevel(logging.DEBUG) # ERROR, WARNING, INFO, or DEBUG (in that order by increasing detail)
+
# improved by Lauren Aldoroty (Duke Univ.)
__author__ = "Lei Hu <leihu@andrew.cmu.edu>"
__version__ = "v1.4"
@@ -72,14 +83,21 @@ def SSS(FITS_obj, FITS_skysub=None, FITS_sky=None, FITS_skyrms=None, FITS_detmas
# * Generate SExtractor OBJECT-MASK
with warnings.catch_warnings():
warnings.simplefilter("ignore")
+ _logger.debug( f"Process {multiprocessing.current_process().pid} running PY_SEx.PS on {FITS_obj}..." )
# NOTE: GAIN, SATURATE, ANALYSIS_THRESH, DEBLEND_MINCONT, BACKPHOTO_TYPE do not affect the detection mask.
- DETECT_MASK = PY_SEx.PS(FITS_obj=FITS_obj, SExParam=['X_IMAGE', 'Y_IMAGE'], GAIN_KEY='PHGAIN', SATUR_KEY=SATUR_KEY, \
- BACK_TYPE='AUTO', BACK_SIZE=BACK_SIZE, BACK_FILTERSIZE=BACK_FILTERSIZE, DETECT_THRESH=DETECT_THRESH, \
- ANALYSIS_THRESH=1.5, DETECT_MINAREA=DETECT_MINAREA, DETECT_MAXAREA=DETECT_MAXAREA, DEBLEND_MINCONT=0.005, \
- BACKPHOTO_TYPE='GLOBAL', CHECKIMAGE_TYPE='OBJECTS', MDIR=MDIR, VERBOSE_LEVEL=VERBOSE_LEVEL)[1][0].astype(bool)
+ DETECT_MASK = PY_SEx.PS(FITS_obj=FITS_obj, SExParam=['X_IMAGE', 'Y_IMAGE'], GAIN_KEY='PHGAIN', SATUR_KEY=SATUR_KEY,
+ BACK_TYPE='AUTO', BACK_SIZE=BACK_SIZE, BACK_FILTERSIZE=BACK_FILTERSIZE,
+ DETECT_THRESH=DETECT_THRESH, ANALYSIS_THRESH=1.5, DETECT_MINAREA=DETECT_MINAREA,
+ DETECT_MAXAREA=DETECT_MAXAREA, DEBLEND_MINCONT=0.005, BACKPHOTO_TYPE='GLOBAL',
+ CHECKIMAGE_TYPE='OBJECTS', MDIR=MDIR, VERBOSE_LEVEL=VERBOSE_LEVEL,
+ logger=_logger
+ )[1][0].astype(bool)
+ _logger.debug( f"...process {multiprocessing.current_process().pid} done running PY_SEx.PS on {FITS_obj}..." )
# * Extract SExtractor SKY-MAP from the Unmasked Image
+ _logger.debug( f"Running fits.getdata({FITS_obj}, ext=0)..." )
PixA_obj = fits.getdata(FITS_obj, ext=0).T
+ _logger.debug( f"...done running fits.getdata({FITS_obj}, ext=0)." )
_PixA = PixA_obj.astype(np.float64, copy=True) # default copy=True, just to emphasize
_PixA[DETECT_MASK] = np.nan
if not _PixA.flags['C_CONTIGUOUS']: _PixA = np.ascontiguousarray(_PixA)
diff --git a/sfft/utils/pyAstroMatic/PYSEx.py b/sfft/utils/pyAstroMatic/PYSEx.py
index c58af2b..875145f 100644
--- a/sfft/utils/pyAstroMatic/PYSEx.py
+++ b/sfft/utils/pyAstroMatic/PYSEx.py
@@ -1,7 +1,12 @@
import re
+import sys
import os
+import io
+import time
import warnings
+import multiprocessing
import subprocess
+import logging
import numpy as np
import os.path as pa
from astropy.io import fits
@@ -16,6 +21,16 @@
__author__ = "Lei Hu <leihu@andrew.cmu.edu>"
__version__ = "v1.4"
+_logger = logging.getLogger('PY_SEx')
+if not _logger.hasHandlers():
+ log_out = logging.StreamHandler(sys.stderr)
+ formatter = logging.Formatter(f'[%(asctime)s - PY_SEx - %(levelname)s] %(message)s', datefmt='%Y-%m-%d %H:%M:%S')
+ log_out.setFormatter(formatter)
+ _logger.addHandler(log_out)
+ _logger.setLevel(logging.DEBUG) # ERROR, WARNING, INFO, or DEBUG (in that order by increasing detail)
+
+
+
class PY_SEx:
@staticmethod
def PS(FITS_obj, PSF_obj=None, FITS_ref=None, SExParam=None, CATALOG_TYPE='FITS_LDAC', \
@@ -25,17 +40,18 @@ def PS(FITS_obj, PSF_obj=None, FITS_ref=None, SExParam=None, CATALOG_TYPE='FITS_
BACKPHOTO_TYPE='LOCAL', PHOT_APERTURES=5.0, NegativeCorr=True, CHECKIMAGE_TYPE='NONE', \
VIGNET=None, STAMP_IMGSIZE=None, AddRD=False, ONLY_FLAGS=None, XBoundary=0.0, YBoundary=0.0, \
Coor4Match='XY_', XY_Quest=None, Match_xytol=2.0, RD_Quest=None, Match_rdtol=1.0, \
- Preserve_NoMatch=False, MDIR=None, VERBOSE_TYPE='QUIET', VERBOSE_LEVEL=2):
+ Preserve_NoMatch=False, MDIR=None, VERBOSE_TYPE='QUIET', VERBOSE_LEVEL=2,
+ logger=_logger):
"""
# Inputs & Outputs:
-FITS_obj [] # FITS file path of the input image for photometry
-
+
-PSF_obj [None] # PSFEx .psf file path for PSF photometry
- -FITS_ref [None] # FITS file path of the input image for detection
- # (a) -FITS_ref = None means single image mode:
+ -FITS_ref [None] # FITS file path of the input image for detection
+ # (a) -FITS_ref = None means single image mode:
# SEx detection & SEx photometry on same image -FITS_obj.
# (b) -FITS_ref != None mean dual image mode:
# SEx detection on -FITS_ref & SEx photometry on -FITS_obj
@@ -57,7 +73,7 @@ def PS(FITS_obj, PSF_obj=None, FITS_ref=None, SExParam=None, CATALOG_TYPE='FITS_
-PIXEL_SCALE [1.0] # SExtractor Parameter PIXEL_SCALE
# size of pixel in arcsec (0=use FITS WCS info)
- # P.S. it only works for surface brightness parameters,
+ # P.S. it only works for surface brightness parameters,
# FWHM_WORLD and star/galaxy separation.
-SEEING_FWHM [1.2] # SExtractor Parameter SEEING_FWHM
@@ -65,7 +81,7 @@ def PS(FITS_obj, PSF_obj=None, FITS_ref=None, SExParam=None, CATALOG_TYPE='FITS_
# P.S. it only works for star/galaxy separation.
-BACK_TYPE ['AUTO'] # SExtractor Parameter BACK_TYPE = [AUTO or MANUAL].
-
+
-BACK_VALUE [0.0] # SExtractor Parameter BACK_VALUE (only work for BACK_TYPE='MANUAL')
-BACK_SIZE [64] # SExtractor Parameter BACK_SIZE
@@ -81,12 +97,12 @@ def PS(FITS_obj, PSF_obj=None, FITS_ref=None, SExParam=None, CATALOG_TYPE='FITS_
-DETECT_MINAREA [5] # SExtractor Parameter DETECT_MINAREA
# min. # of pixels above threshold
-
+
-DETECT_MAXAREA [0] # SExtractor Parameter DETECT_MAXAREA
# max. # of pixels above threshold (0=unlimited)
-DEBLEND_MINCONT [0.005] # SExtractor Parameter DEBLEND_MINCONT (typically, 0.001 - 0.005)
- # Minimum contrast parameter for deblending
+ # Minimum contrast parameter for deblending
-CLEAN ['Y] # SExtractor Parameter CLEAN
# Clean spurious detections? (Y or N)?
@@ -100,7 +116,7 @@ def PS(FITS_obj, PSF_obj=None, FITS_ref=None, SExParam=None, CATALOG_TYPE='FITS_
# P.S. Here it can be a Python list of apertures
-CHECKIMAGE_TYPE ['NONE'] # SExtractor Parameter CHECKIMAGE_TYPE
- # can be NONE, BACKGROUND, BACKGROUND_RMS, MINIBACKGROUND, MINIBACK_RMS,
+ # can be NONE, BACKGROUND, BACKGROUND_RMS, MINIBACKGROUND, MINIBACK_RMS,
# -BACKGROUND, FILTERED, OBJECTS, -OBJECTS, SEGMENTATION, or APERTURES
-VERBOSE_TYPE ['QUIET'] # SExtractor Parameter VERBOSE_TYPE
@@ -108,12 +124,12 @@ def PS(FITS_obj, PSF_obj=None, FITS_ref=None, SExParam=None, CATALOG_TYPE='FITS_
# Other parameters
- -NegativeCorr [True] # In SExtractor, MAG_* = 99. and MAGERR_* = 99. for FLUX_* < 0.0
- # If -NegativeCorr = True, PYSEx will correct MAG_* and MAGERR_*
+ -NegativeCorr [True] # In SExtractor, MAG_* = 99. and MAGERR_* = 99. for FLUX_* < 0.0
+ # If -NegativeCorr = True, PYSEx will correct MAG_* and MAGERR_*
# to be a valid values using abs(FLUX_*) and FLUXERR_*.
-VIGNET [None] # VIGNET for generating PSFEx input catalog
- # e.g., set -VIGNET = (51, 51), PYSEx will add 'VIGNET(51, 51)' into
+ # e.g., set -VIGNET = (51, 51), PYSEx will add 'VIGNET(51, 51)' into
# the SExtractor output parameter list.
-STAMP_IMGSIZE [None] # PYSEx allows for making a stamp for each detected source on -FITS_obj,
@@ -122,14 +138,14 @@ def PS(FITS_obj, PSF_obj=None, FITS_ref=None, SExParam=None, CATALOG_TYPE='FITS_
-AddRD [False] # Add columns for Ra. and Decl. in the output catalog?
# P.S. The columns are X_WORLD, Y_WORLD or XWIN_WORLD, YWIN_WORLD.
- # Although SExtractor itself can generate these columns, here we use
+ # Although SExtractor itself can generate these columns, here we use
# astropy.wcs to convert image coordinates to world coordinates instead.
# (Because I feel like that astropy has better WCS compatibility than SExtractor)
-ONLY_FLAGS [None] # Do you put any constrain on the SExtractor output parameter FLAGS
#
# FLAGS description
- # 1 aperture photometry is likely to be biased by neighboring sources
+ # 1 aperture photometry is likely to be biased by neighboring sources
# or by more than 10% of bad pixels in any aperture
# 2 the object has been deblended
# 4 at least one object pixel is saturated
@@ -164,7 +180,7 @@ def PS(FITS_obj, PSF_obj=None, FITS_ref=None, SExParam=None, CATALOG_TYPE='FITS_
-Preserve_NoMatch [False] # Preserve the detected sources in SExtractor photometry catalog without cross match counterpart
-MDIR [None] # Parent Directory for output files
- # PYSEx will generate a child directory with a random name under the paraent directory
+ # PYSEx will generate a child directory with a random name under the paraent directory
# all output files are stored in the child directory
-VERBOSE_LEVEL [2] # The level of verbosity, can be [0, 1, 2]
@@ -177,34 +193,34 @@ def PS(FITS_obj, PSF_obj=None, FITS_ref=None, SExParam=None, CATALOG_TYPE='FITS_
PixA_SExCheckLst # List of Pixel arrays for SExtractor check images
# P.S. PixA = fits.getdata(FITS, ext=0).T
-
+
FITS_SExCat # File path of SExtractor photometry catalog
# P.S. only for -MDIR is not None
FITS_SExCheckLst # List of file path of SExtractor check images
# P.S. only for -MDIR is not None
-
- # ---------------- MORE DESCRIPTION ON HOW SEXTRACTOR WORK ----------------
+
+ # ---------------- MORE DESCRIPTION ON HOW SEXTRACTOR WORK ----------------
#
# * SExtractor Inputs
# ** Array-Inputs:
- # SEx works on one image for signal detection and another image for photometry
+ # SEx works on one image for signal detection and another image for photometry
# @ Individual Mode (Common): Image4detect and Image4phot are the same image (-FITS_obj).
# @ Dual Mode: Image4detect (-FITS_ref) and Image4phot (-FITS_obj) are different images .
# ** PSF-Input:
# SEx can accept given PSF model for PSF-Photometry (-PSF_obj).
- # ** Parameter-Inputs:
+ # ** Parameter-Inputs:
# a. Basic keywords in FITS header of Image4detect:
# (-GAIN_KEY, -SATUR_KEY).
# b. How to generate Global Background Map:
# (-BACK_TYPE, -BACK_VALUE, -BACK_SIZE, -BACK_FILTERSIZE).
# c. Give the criteria for SExtractor Source Detection
# (-DETECT_THRESH, -DETECT_MINAREA, -DETECT_MAXAREA, -DEBLEND_NTHRESH, -DEBLEND_MINCONT, -CLEAN).
- # d. Which photometry method(s) used by SExtractor:
+ # d. Which photometry method(s) used by SExtractor:
# (parameters in -SExParam, e.g., FLUX_AUTO, FLUX_APER, FLUX_PSF).
- # e. Specify output Check-Images:
+ # e. Specify output Check-Images:
# (-CHECKIMAGE_TYPE).
- # f. Specify output columns in output SExtractor photometry table:
+ # f. Specify output columns in output SExtractor photometry table:
# (-SExParam).
#
# Remarks on Weight-Map:
@@ -212,50 +228,50 @@ def PS(FITS_obj, PSF_obj=None, FITS_ref=None, SExParam=None, CATALOG_TYPE='FITS_
# vairable effective GAIN and background noise level across the field due to different Num_Exposure.
# Weight-map would help SExtractor to calculate errors, such as, FLUXERR and MAGERR, more accurately.
#
- # WARNING: For current version, PYSEx does not include this feature, and I would recommend users
- # to set an average effective GAIN for mosaic image, and keep in mind it may, to some extent,
+ # WARNING: For current version, PYSEx does not include this feature, and I would recommend users
+ # to set an average effective GAIN for mosaic image, and keep in mind it may, to some extent,
# cause inaccurate error estimations.
#
# * SExtractor Workflow (Background)
# ** Extract Global_Background_Map (GBMap) and its RMS (GBRMap) from Image4detect & Image4phot
# Control Parameters: BACK_TYPE, BACK_VALUE, BACK_SIZE and BACK_FILTERSIZE.
- # @ Produce GBMap
+ # @ Produce GBMap
# a. Manual-FLAT (e.g. BACK_TYPE='MANUAL', BACK_VALUE=100.0)
# SEx directly define GBMap as a FLAT image with given constant BACK_VALUE.
# b. Auto (e.g. BACK_TYPE='AUTO', BACK_SIZE=64, BACK_FILTERSIZE=3)
# SEx defines a mesh of a grid that covers the whole frame by [BACK_SIZE].
# i. Convergence-based sigma-clipping method on the flux histogram of each tile.
- # More specificly, the flux histogram is clipped iteratively until convergence
+ # More specificly, the flux histogram is clipped iteratively until convergence
# at +/- 3sigma around its median.
# ii. SEx compute local background estimator from the clipped histogram of each tile.
- # If sigma is changed by less than 20% during clipping process (the tile is not crowded),
+ # If sigma is changed by less than 20% during clipping process (the tile is not crowded),
# use the mean of the clipped histogram as estimator
# otherwise (the tile is crowded), mode = 2.5*median - 1.5*mean is employed instead.
- # iii. Once the estimate grid is calculated, a median filter [BACK_FILTERSIZE] can be applied
+ # iii. Once the estimate grid is calculated, a median filter [BACK_FILTERSIZE] can be applied
# to suppress possible local overestimations.
- # iv. The resulting background map is them simply a bicubic-spline interpolation
+ # iv. The resulting background map is them simply a bicubic-spline interpolation
# between the meshes of the grid.
#
# @ Generate GBRMap
# only Auto (e.g, BACK_SIZE=64, BACK_FILTERSIZE=3)
- # SEx produces the noise map by the same approach of Auto style of GBMap, where the only
- # difference is that SEx is [probably] use standard deviation as estimator of the
+ # SEx produces the noise map by the same approach of Auto style of GBMap, where the only
+ # difference is that SEx is [probably] use standard deviation as estimator of the
# clipped flux historgram, other than mean or mode.
- #
+ #
# NOTE Abbr. GBMap / GBRMap from Image4detect: GBMap_4d / GBRMap_4d
# Abbr. GBMap / GBRMap from Image4phot: GBMap_4p / GBRMap_4p
#
- # NOTE WARNING: Dual Mode have to use consistent control parameters for Image4detect & Image4phot,
- # as it is not allowed to set some secondary configuration in SEx software framework,
+ # NOTE WARNING: Dual Mode have to use consistent control parameters for Image4detect & Image4phot,
+ # as it is not allowed to set some secondary configuration in SEx software framework,
# despite that it is not necessarily reasonable in some cases.
#
# * SExtractor Workflow (Detection)
# ** a. SEx-Detect on Image4detect: SkySubtraction & Filtering & Thresholding & Deblending & AreaConstrain & Clean
# @ SkySubtraction & Filtering Process (e.g. FILTER='Y', FILTER_NAME='default.conv')
- # SEx Remove GBMap_4d from Image4detect and then perform a convolution to maximizes detectability.
+ # SEx Remove GBMap_4d from Image4detect and then perform a convolution to maximizes detectability.
# NOTE The power-spectrum of the noise and that of the superimposed signal can be significantly different.
# NOTE Although Filtering is a benefit for detection, it distorts profiles and correlates the noise.
- # NOTE Filtering is applied 'on the fly' to the image, and directly affects only the following
+ # NOTE Filtering is applied 'on the fly' to the image, and directly affects only the following
# Thresholding process and Isophotal parameters.
#
# @ Thresholding Process (e.g. DETECT_THRESH=1.5, THRESH_TYPE='RELATIVE')
@@ -263,20 +279,20 @@ def PS(FITS_obj, PSF_obj=None, FITS_ref=None, SExParam=None, CATALOG_TYPE='FITS_
# We would be bettter to imagine SEx actually make a hovering mask.
#
# @ Deblending Process (e.g. DEBLEND_MINCONT=0.005, DEBLEND_NTHRESH=32)
- # SEx triggers a deblending process to dentify signal islands from the hovering
+ # SEx triggers a deblending process to dentify signal islands from the hovering
# mask [probably] on Filtered_Image,
# which converts the hovering mask to be a hovering label map.
#
# @ Put AreaConstrain (e.g. DETECT_MINAREA=5, DETECT_MAXAREA=0)
- # MinMax AreaContrain is applied and then iolated cases then lose their hovering labels.
+ # MinMax AreaContrain is applied and then iolated cases then lose their hovering labels.
#
# @ Clean Process (e.g. CLEAN='YES')
# SEx will clean the list of objects of artifacts caused by bright objects.
# As a correction process, all cleaned objects are subsequently removed from the hovering label map.
- # NOTE Now the hovering label map is the SEGMENTATION check image.
+ # NOTE Now the hovering label map is the SEGMENTATION check image.
# One may refer to such label island as ISOIsland
# NOTE These labels are consistent with the indices in the ouput photometry table.
- # NOTE SEx will report something like this: Objects: detected 514 / sextracted 397
+ # NOTE SEx will report something like this: Objects: detected 514 / sextracted 397
# SET CLEAN='N', you could find detected == sextracted.
#
# ** b. Generate Positional & BasicShape Paramters from isophotal profile
@@ -289,7 +305,7 @@ def PS(FITS_obj, PSF_obj=None, FITS_ref=None, SExParam=None, CATALOG_TYPE='FITS_
# (if a parameter can be fully expressed by them, we will use $ to indicate).
# iv. Describe ISOIsand as an elliptical shape, centred at Barycenter.
# $A_IMAGE, $B_IMAGE are ellipse semi-major and semi-minor axis lengths, respectively.
- # The ellipse is uniquely determined by $CXX_IMAGE, $CYY_IMAGE, $CXY_IMAGE with KRON_RADIUS,
+ # The ellipse is uniquely determined by $CXX_IMAGE, $CYY_IMAGE, $CXY_IMAGE with KRON_RADIUS,
# where KRON_RADIUS is independently calculated in a routine inspired by Kron's 'first moment' algorithm.
# NOTE By-products: $ELONGATION = A / B and $ELLIPTICITY = 1 - B / A
#
@@ -300,19 +316,19 @@ def PS(FITS_obj, PSF_obj=None, FITS_ref=None, SExParam=None, CATALOG_TYPE='FITS_
# METHOD: The computations involved are roughly the same except that the domain is a circular Gaussian window,
# (I don't know what is the window radius) as opposed to the object's isophotal footprint (ISOIsland).
#
- # MOTIVATION: Parameters measured within an object's isophotal limit are sensitive to two main factors:
- # + Changes in the detection threshold, which create a variable bias
- # + Irregularities in the object's isophotal boundaries, which act as
+ # MOTIVATION: Parameters measured within an object's isophotal limit are sensitive to two main factors:
+ # + Changes in the detection threshold, which create a variable bias
+ # + Irregularities in the object's isophotal boundaries, which act as
# additional 'noise' in the measurements.
#
- # @Positional: This is an iterative process. The computation starts by initializing
+ # @Positional: This is an iterative process. The computation starts by initializing
# the windowed centroid coordinates to the Barycenter.
- # The process will adjust window and finally its centroid converges
+ # The process will adjust window and finally its centroid converges
# at some point: XWIN_IMAGE, YWIN_IMAGE.
# (If the process is failed then XWIN_IMAGE, YWIN_IMAGE = X_IMAGE, Y_IMAGE)
- # It has been verified that for isolated, Gaussian-like PSFs, its accuracy is close to
- # the theoretical limit set by image noise.
- # NOTE: We preferably use it for point sources, like in transient detection.
+ # It has been verified that for isolated, Gaussian-like PSFs, its accuracy is close to
+ # the theoretical limit set by image noise.
+ # NOTE: We preferably use it for point sources, like in transient detection.
# However it may not optimal for extended sources like glaxies.
# NOTE: X_IMAGE & Y_IMAGE seems to be a good compromise choice.
#
@@ -320,9 +336,9 @@ def PS(FITS_obj, PSF_obj=None, FITS_ref=None, SExParam=None, CATALOG_TYPE='FITS_
# X2WIN_IMAGE, Y2WIN_IMAGE, XY2WIN_IMAGE
# AWIN_IMAGE, BWIN_IMAGE, CXXWIN_IMAGE, CYYWIN_IMAGE, CXYWIN_IMAGE
#
- # NOTE: Positional XWIN_IMAGE and YWIN_IMAGE are quite useful to
- # provide a refined Object coordinate (of gaussian-like point sources).
- # However we seldom use Window version of BasicShape parameters
+ # NOTE: Positional XWIN_IMAGE and YWIN_IMAGE are quite useful to
+ # provide a refined Object coordinate (of gaussian-like point sources).
+ # However we seldom use Window version of BasicShape parameters
# to describe the shape of the approaximated ellipse.
#
# ** d. Generate Positional Paramters from PSF Fitting
@@ -334,13 +350,13 @@ def PS(FITS_obj, PSF_obj=None, FITS_ref=None, SExParam=None, CATALOG_TYPE='FITS_
# NOTE This process always works on Image4phot.
# @ISO: SEx simply count flux according to the hovering label map (ISOIsland).
# @APER: SEx count flux on a Circular Aperture with given PHOT_APERTURES (centred at Barycenter X_IMAGE, Y_IMAGE).
- # @AUTO: SEx count flux on a Elliptic Aperture, which is determined by the hovering isophotal ellipse
+ # @AUTO: SEx count flux on a Elliptic Aperture, which is determined by the hovering isophotal ellipse
# (centred at Barycenter X_IMAGE, Y_IMAGE). The leaked light fraction is typically less than 10%.
# @PSF: SEx count flux according to the PSF fitting results (centred at XPSF_IMAGE and YPSF_IMAGE).
# This is optimal for pointsource but fairly wrong for extended objects.
#
# ** Peel background contribution from Counted Flux
- # @BACKPHOTO_TYPE='LOCAL': background will take a rectangular annulus into account,
+ # @BACKPHOTO_TYPE='LOCAL': background will take a rectangular annulus into account,
# which has a donnut shape around the object, measured on Image4phot.
# @BACKPHOTO_TYPE='GLOBAL': background will directly use GBMap_4p,
# which can be Manual-Flat or Auto.
@@ -356,54 +372,54 @@ def PS(FITS_obj, PSF_obj=None, FITS_ref=None, SExParam=None, CATALOG_TYPE='FITS_
# @BACKGROUND : GBMap_4p
# @BACKGROUND_RMS : GBRMap_4d
# @-BACKGROUND : Image4phot - GBMap_4p
- # @FILTERED: Filtered_Image
+ # @FILTERED: Filtered_Image
# @SEGMENTATION: ISOIsland Label Map
# @OBJECTS: ISOIslands use flux in -BACKGROUND, zero otherwise
- # @APERTURES: -BACKGROUND with brighten apertures to show the isophotal ellipses.
+ # @APERTURES: -BACKGROUND with brighten apertures to show the isophotal ellipses.
#
# * SExtractor Workflow (Formula)
# a. MAG = -2.5 * log10(FLUX) + MAG_ZEROPOINT
- # b. MAGERR = 1.0857 * FLUXERR / FLUX = 1.0857 / SNR
+ # b. MAGERR = 1.0857 * FLUXERR / FLUX = 1.0857 / SNR
# NOTE: a common crude approxmation SNR ~ 1/MAGERR.
- # NOTE: It is [probably] derived from MAGERR = 2.5*np.log10(1.0+1.0/SNR)
+ # NOTE: It is [probably] derived from MAGERR = 2.5*np.log10(1.0+1.0/SNR)
# Ref: http://www.ucolick.org/~bolte/AY257/s_n.pdf
# ZTF use a more precise factor 1.085736, whatever, not a big deal.
#
# * Additional Clarification
# a. SExtractor can read GAIN & SATURATION & MAGZERO from FITS header of Image4phot by their keys.
- # b. If SExtractor is configured with FITS_LDAC, the FITS header of Image4phot will be delivered
- # into output FITS file saved at output-FITS[1].data in table format
+ # b. If SExtractor is configured with FITS_LDAC, the FITS header of Image4phot will be delivered
+ # into output FITS file saved at output-FITS[1].data in table format
# (only one element: a long integrated header text).
- # c. If some ISOIsland has saturated pixel value on Image4phot, you can still find it
+ # c. If some ISOIsland has saturated pixel value on Image4phot, you can still find it
# on SEGMENTATION / OBJECTS, and it will be marked by FLAGS=4 in output catalog.
# d. Windowed Coordinate has higher priority in the function
# i. Make Stamps ii. Convert to RD iii. Symmetric-Match
- # First use XWIN_IMAGE & YWIN_IMAGE (if exist),
+ # First use XWIN_IMAGE & YWIN_IMAGE (if exist),
# otherwise, employ X_IMAGE & Y_IMAGE instead.
- # e. SExtractor allows to submit request for multi-check images
+ # e. SExtractor allows to submit request for multi-check images
# e.g. CHECKIMAGE_TYPE = "BACKGROUND,SEGMENTATION,..."
# f. Although SExtractor can directly provide sky coordinates in output table,
- # We always independently to get them by convert XY using astropy.
- # g. If VIGNET is called in SExtractor, stamps will be extracted around the targets,
+ # We always independently to get them by convert XY using astropy.
+ # g. If VIGNET is called in SExtractor, stamps will be extracted around the targets,
# centred at their Barycenter X_IMAGE, Y_IMAGE, from Image4phot.
#
# * Additional Tips
# a. SNR_WIN: Window-based Gaussian-weighted SNR estimate
- # Although SNR_WIN is empirically tend to slightly underestimate noise,
+ # Although SNR_WIN is empirically tend to slightly underestimate noise,
# this useful parameter is calculated independently from the employed phot-method.
#
- # b. If you got a long runtime, it may caused by
- # i. Low DETECT_THRESH
+ # b. If you got a long runtime, it may caused by
+ # i. Low DETECT_THRESH
# ii. Request XWIN_IMAGE, YWIN_IMAGE, SNR_WIN
# iii. BACKPHOTO_TYPE == 'LOCAL'
#
# c. WARNINGS
- # i. You may encounter bug if sethead after fits.getdata
+ # i. You may encounter bug if sethead after fits.getdata
# ii. SExtractor do not support string > 256 as argument in command line.
# Use 'cd dir && sex ...' to avoid segmentation fault.
# iii. FITS_LDAC --- TABLE-HDU 2 | FITS_1.0 --- TABLE-HDU 1
# iv. In Debian operation system, please correct 'sex' as 'sextractor'
- #
+ #
# d. A coarse look-up table between DETECT_THRESH and MINIMAL SNR_WIN
# DETECT_THRESH = 1.0 ---> minimal SNR_WIN = 3-4
# DETECT_THRESH = 1.2 ---> minimal SNR_WIN = 4-5
@@ -419,46 +435,49 @@ def PS(FITS_obj, PSF_obj=None, FITS_ref=None, SExParam=None, CATALOG_TYPE='FITS_
# ** When this mode is called, please read above comments very carefully.
#
# @ Single-Image Mode: FITS_obj is Image4detect & Image4phot.
- # ** When sky background has been well subtracted.
+ # ** When sky background has been well subtracted.
# Typically Set: BACK_TYPE='MANUAL', BACK_VALUE=0.0, BACK_SIZE=64, BACK_FILTERSIZE=3, BACKPHOTO_TYPE='LOCAL'
# a. BACK_TYPE & BACK_VALUE: Use Flat-Zero as Global_Background_Map
# b. BACK_SIZE & BACK_FILTERSIZE: Produce RMS of Global_Background_Map [AUTO]
# c. BACKPHOTO_TYPE: Use LOCAL / GLOBAL (zero) background to count sky flux contribution in photometry.
#
# * Additional Remarks on Background
- # @ well-subtracted sky just means it probably outperforms SEx GLOBAL-Sky,
- # we should not presume the underlying true sky is really subtracted, therefore,
+ # @ well-subtracted sky just means it probably outperforms SEx GLOBAL-Sky,
+ # we should not presume the underlying true sky is really subtracted, therefore,
# BACKPHOTO_TYPE = 'LOCAL' is in general necessary !
#
- # @ Just like sky subtraction, the sky term in image subtraction can only handle the
- # low-spatial-frequency trend of background. The image has flat zero background as an ideal expectation,
+ # @ Just like sky subtraction, the sky term in image subtraction can only handle the
+ # low-spatial-frequency trend of background. The image has flat zero background as an ideal expectation,
# we still need set BACKPHOTO_TYPE = 'LOCAL' on difference photometry.
#
- # @ 'LOCAL' method itself can be biased too. We get the conclusion when we try to perform
- # aperture photometry on two psf-homogenized DECam images. Recall any error on matched kernel is some
- # linear effect for aperture photometry, however, we found a bias in FLUX_APER which is independent
- # with the target birghtness. E.g. FLUX_APER_SCI is always smaller than FLUX_APER_REF with
- # a nearly constant value, say 10.0 ADU, no matter the target is 19.0 mag or 22.0 mag, as if there is
- # some constant leaked light when we measure on SCI. Equivalently, DMAG = MAG_APER_SCI - MAG_APER_REF
+ # @ 'LOCAL' method itself can be biased too. We get the conclusion when we try to perform
+ # aperture photometry on two psf-homogenized DECam images. Recall any error on matched kernel is some
+ # linear effect for aperture photometry, however, we found a bias in FLUX_APER which is independent
+ # with the target birghtness. E.g. FLUX_APER_SCI is always smaller than FLUX_APER_REF with
+ # a nearly constant value, say 10.0 ADU, no matter the target is 19.0 mag or 22.0 mag, as if there is
+ # some constant leaked light when we measure on SCI. Equivalently, DMAG = MAG_APER_SCI - MAG_APER_REF
# deviate zero-baseline, and it becomes increasingly serious (towards to the faint end).
#
- # ------------------------------------
- # We guess the problem is caused by the fact: the background value calculated from the annulus
- # around target might be a biased (over/under-) estimation. One observation supports our argument:
+ # ------------------------------------
+ # We guess the problem is caused by the fact: the background value calculated from the annulus
+ # around target might be a biased (over/under-) estimation. One observation supports our argument:
# the flux bias is much more evident when we increase the aperture size.
- # NOTE: As we have found the flux bias is basically a constant, we can do relative-calibration by
- # calculating the compensation offset FLUX_APER_REF - FLUX_APER_SCI for a collection of sationary stars.
- # It is 'relative' since we have just assumed background estimation of REF is correct,
+ # NOTE: As we have found the flux bias is basically a constant, we can do relative-calibration by
+ # calculating the compensation offset FLUX_APER_REF - FLUX_APER_SCI for a collection of sationary stars.
+ # It is 'relative' since we have just assumed background estimation of REF is correct,
# which is proper if we are going to derive the variability (light curve).
#
# @ In which cases, Re-Run SExtractor can get the same coordinate list?
- # a. Same configurations but only change photometric method, e.g. from AUTO to APER
+ # a. Same configurations but only change photometric method, e.g. from AUTO to APER
# b. Same configurations but from Single-Image Mode to Dual-Image Mode
# NOTE: FLAGS is correlated to object image, if we add constraint FLAG=0
# then we fail to get the same coordinate list.
#
"""
+ pid = multiprocessing.current_process().pid
+ logger.debug( f"PY_SEx.PS process {pid} starting {FITS_obj}" )
+
# * sex or sextractor?
for cmd in ['sex', 'sextractor']:
try:
@@ -479,34 +498,60 @@ def PS(FITS_obj, PSF_obj=None, FITS_ref=None, SExParam=None, CATALOG_TYPE='FITS_
_message = 'Run Python Wrapper of SExtractor!'
print('\nMeLOn CheckPoint [%s]: %s' %(objname, _message))
- # * Keyword Configurations
- phr_obj = fits.getheader(FITS_obj, ext=0)
+ # * Keyword Configurations
+ logger.debug( f"PY_SEx process {pid} reading header from {FITS_obj}..." )
+ retries = 5
+ while retries > 0:
+ # NOTE -- we were getting weird race condition issues where sometimes
+ # the FITS header reading failed. We aren't sure exactly what's going
+ # on, but the hypothesis is bad filesystem behavior. The earlier
+ # writefits returns, but somehow the file wasn't really fully written
+ # to either disk, or to the memory buffers that the fits.getheader
+ # here read. Put in an ugly hack of retring and sleeping to work
+ # around this.
+ try:
+ phr_obj = fits.getheader(FITS_obj, ext=0)
+ retries = 0
+ except Exception as ex:
+ retries -= 1
+ strio = io.StringIO()
+ strio.write( f"PY_SEx process {pid} got exception trying to read fits header: {ex}; " )
+ if retries > 0:
+ strio.write( f"waiting 1s and trying again ({retries} tries left)" )
+ logger.warning( strio.getvalue() )
+ time.sleep( 1 )
+ else:
+ strio.write( "repeatedly failed, giving up" )
+ logger.error( strio.getvalue() )
+ raise RuntimeError( f"Failed to read header from {FITS_obj}" )
+
+ logger.debug( f"... DONE: PY_SEx process {pid} reading header from {FITS_obj}" )
if GAIN_KEY in phr_obj:
GAIN = phr_obj[GAIN_KEY]
if VERBOSE_LEVEL in [1, 2]:
_message = 'SExtractor uses GAIN = [%s] from keyword [%s]!' %(GAIN, GAIN_KEY)
print('MeLOn CheckPoint [%s]: %s' %(objname, _message))
- else:
+ else:
GAIN = 0.0 # infinite GAIN, Poission noise ignored
if VERBOSE_LEVEL in [0, 1, 2]:
_warn_message = 'SExtractor has to use default GAIN = 0!'
warnings.warn('MeLOn WARNING [%s]: %s' %(objname, _warn_message))
-
+
if SATUR_KEY in phr_obj:
SATURATION = phr_obj[SATUR_KEY]
if VERBOSE_LEVEL in [1, 2]:
_message = 'SExtractor uses SATURATION = [%s] from keyword [%s]!' %(SATURATION, SATUR_KEY)
print('MeLOn CheckPoint [%s]: %s' %(objname, _message))
- else:
+ else:
SATURATION = 50000.0
if VERBOSE_LEVEL in [0, 1, 2]:
_warn_message = 'SExtractor has to use default SATURATION = 50000.0!'
warnings.warn('MeLOn WARNING [%s]: %s' %(objname, _warn_message))
-
+
"""
# A few additional remarks
# [1] MAGERR/FLUXERR/SNR are very sensitive to GAIN value.
- # [2] PIXEL_SCALE (unit: arcsec) only works for surface brightness parameters,
+ # [2] PIXEL_SCALE (unit: arcsec) only works for surface brightness parameters,
# FWHM (FWHM_WORLD) and star/galaxy separation. PIXEL_SCALE=0 uses FITS WCS info.
# [3] SEEING_FWHM (unit: arcsec) is only for star/galaxy separation.
# VIP: You'd better to give a good estimate if star/galaxy separation is needed.
@@ -522,10 +567,10 @@ def PS(FITS_obj, PSF_obj=None, FITS_ref=None, SExParam=None, CATALOG_TYPE='FITS_
ConfigDict['GAIN_KEY'] = '%s' %GAIN_KEY
ConfigDict['SATUR_KEY'] = '%s' %SATUR_KEY
- ConfigDict['MAG_ZEROPOINT'] = '0.0'
+ ConfigDict['MAG_ZEROPOINT'] = '0.0'
ConfigDict['PIXEL_SCALE'] = '%s' %PIXEL_SCALE
ConfigDict['SEEING_FWHM'] = '%s' %SEEING_FWHM
-
+
ConfigDict['BACK_TYPE'] = '%s' %BACK_TYPE
ConfigDict['BACK_VALUE'] = '%s' %BACK_VALUE
ConfigDict['BACK_SIZE'] = '%s' %BACK_SIZE
@@ -538,14 +583,14 @@ def PS(FITS_obj, PSF_obj=None, FITS_ref=None, SExParam=None, CATALOG_TYPE='FITS_
ConfigDict['DEBLEND_NTHRESH'] = '%s' %DEBLEND_NTHRESH
ConfigDict['DEBLEND_MINCONT'] = '%s' %DEBLEND_MINCONT
ConfigDict['CLEAN'] = '%s' %CLEAN
-
+
ConfigDict['CHECKIMAGE_TYPE'] = '%s' %CHECKIMAGE_TYPE
ConfigDict['BACKPHOTO_TYPE'] = '%s' %BACKPHOTO_TYPE
if not isinstance(PHOT_APERTURES, (int, float)):
ConfigDict['PHOT_APERTURES'] = '%s' %(','.join(np.array(PHOT_APERTURES).astype(str)))
else: ConfigDict['PHOT_APERTURES'] = '%s' %PHOT_APERTURES
if PSF_obj is not None: ConfigDict['PSF_NAME'] = '%s' %PSF_obj
-
+
# create configuration file .conv
if USE_FILT:
# see https://github.com/astromatic/sextractor/blob/master/config/default.conv
@@ -558,11 +603,11 @@ def PS(FITS_obj, PSF_obj=None, FITS_ref=None, SExParam=None, CATALOG_TYPE='FITS_
_cfile = open(conv_path, 'w')
_cfile.write(conv_text)
_cfile.close()
-
+
USE_NNW = False
- if 'CLASS_STAR' in SExParam:
+ if 'CLASS_STAR' in SExParam:
USE_NNW = True
-
+
if USE_NNW:
# see https://github.com/astromatic/sextractor/blob/master/config/default.nnw
nnw_text = r"""
@@ -572,12 +617,12 @@ def PS(FITS_obj, PSF_obj=None, FITS_ref=None, SExParam=None, CATALOG_TYPE='FITS_
# outputs: ``Stellarity index'' (0.0 to 1.0)
# Seeing FWHM range: from 0.025 to 5.5'' (images must have 1.5 < FWHM < 5 pixels)
# Optimized for Moffat profiles with 2<= beta <= 4.
-
+
3 10 10 1
-
+
-1.56604e+00 -2.48265e+00 -1.44564e+00 -1.24675e+00 -9.44913e-01 -5.22453e-01 4.61342e-02 8.31957e-01 2.15505e+00 2.64769e-01
3.03477e+00 2.69561e+00 3.16188e+00 3.34497e+00 3.51885e+00 3.65570e+00 3.74856e+00 3.84541e+00 4.22811e+00 3.27734e+00
-
+
-3.22480e-01 -2.12804e+00 6.50750e-01 -1.11242e+00 -1.40683e+00 -1.55944e+00 -1.84558e+00 -1.18946e-01 5.52395e-01 -4.36564e-01 -5.30052e+00
4.62594e-01 -3.29127e+00 1.10950e+00 -6.01857e-01 1.29492e-01 1.42290e+00 2.90741e+00 2.44058e+00 -9.19118e-01 8.42851e-01 -4.69824e+00
-2.57424e+00 8.96469e-01 8.34775e-01 2.18845e+00 2.46526e+00 8.60878e-02 -6.88080e-01 -1.33623e-02 9.30403e-02 1.64942e+00 -1.01231e+00
@@ -588,11 +633,11 @@ def PS(FITS_obj, PSF_obj=None, FITS_ref=None, SExParam=None, CATALOG_TYPE='FITS_
3.75075e+00 7.25399e+00 -1.75325e+00 -2.68814e+00 -3.71128e+00 -4.62933e+00 -2.13747e+00 -1.89186e-01 1.29122e+00 -7.49380e-01 6.71712e-01
-8.41923e-01 4.64997e+00 5.65808e-01 -3.08277e-01 -1.01687e+00 1.73127e-01 -8.92130e-01 1.89044e+00 -2.75543e-01 -7.72828e-01 5.36745e-01
-3.65598e+00 7.56997e+00 -3.76373e+00 -1.74542e+00 -1.37540e-01 -5.55400e-01 -1.59195e-01 1.27910e-01 1.91906e+00 1.42119e+00 -4.35502e+00
-
+
-1.70059e+00 -3.65695e+00 1.22367e+00 -5.74367e-01 -3.29571e+00 2.46316e+00 5.22353e+00 2.42038e+00 1.22919e+00 -9.22250e-01 -2.32028e+00
-
-
- 0.00000e+00
+
+
+ 0.00000e+00
1.00000e+00 """
nintent = len(re.split('NNW', re.split('\n', nnw_text)[1])[0])
@@ -605,9 +650,9 @@ def PS(FITS_obj, PSF_obj=None, FITS_ref=None, SExParam=None, CATALOG_TYPE='FITS_
# create configuration file .param
USExParam = SExParam.copy()
if SExParam is None: USExParam = []
- if 'X_IMAGE' not in USExParam:
+ if 'X_IMAGE' not in USExParam:
USExParam.append('X_IMAGE')
- if 'Y_IMAGE' not in USExParam:
+ if 'Y_IMAGE' not in USExParam:
USExParam.append('Y_IMAGE')
Param_path = ''.join([TDIR, "/PYSEx.param"])
@@ -630,19 +675,36 @@ def PS(FITS_obj, PSF_obj=None, FITS_ref=None, SExParam=None, CATALOG_TYPE='FITS_
sex_config_path = AMConfig_Maker.AMCM(MDIR=TDIR, AstroMatic_KEY=SEx_KEY, \
ConfigDict=ConfigDict, tag='PYSEx')
-
+
# * Trigger SExtractor
if FITS_ref is None:
- os.system("cd %s && %s %s -c %s -CATALOG_NAME %s -CHECKIMAGE_NAME %s" \
- %(pa.dirname(FITS_obj), SEx_KEY, FNAME, sex_config_path, \
- FITS_SExCat, ','.join(FITS_SExCheckLst)))
-
+ cmd_to_run = [ SEx_KEY, FNAME, '-c', sex_config_path,
+ '-CATALOG_NAME', FITS_SExCat,
+ '-CHECKIMAGE_NAME', ','.join(FITS_SExCheckLst) ]
+ logger.debug( f"PY_SEx process {pid} running: {cmd_to_run}..." )
+ res = subprocess.run( cmd_to_run,
+ cwd=pa.dirname(FITS_obj),
+ capture_output=True )
+ if res.returncode != 0:
+ barf = ( f"*** PY_SEx.PS process {pid} {SEx_KEY} subprocess return code {res.returncode}\n"
+ f"*** stdout ***\n{res.stdout}\n*** stderr ***\n{res.stderr}\n" )
+ logger.error( barf )
+ raise RuntimeError( f"{SEx_KEY} failed: {barf}" )
+ logger.debug( f"...DONE: PY_SEx process {pid} running: {cmd_to_run}" )
+
+ # os.system("cd %s && %s %s -c %s -CATALOG_NAME %s -CHECKIMAGE_NAME %s" \
+ # %(pa.dirname(FITS_obj), SEx_KEY, FNAME, sex_config_path, \
+ # FITS_SExCat, ','.join(FITS_SExCheckLst)))
+
if FITS_ref is not None:
# WARNING: more risky to fail due to the too long string.
+
+ raise RuntimeError( "FITS_ref was not None, fix the code." )
+
os.system("%s %s,%s -c %s -CATALOG_NAME %s -CHECKIMAGE_NAME %s" \
%(SEx_KEY, FITS_ref, FITS_obj, sex_config_path, \
- FITS_SExCat, ','.join(FITS_SExCheckLst)))
-
+ FITS_SExCat, ','.join(FITS_SExCheckLst)))
+
"""
# Deprecated as it requires WCSTools, and seems not very useful
def record(FITS):
@@ -653,7 +715,7 @@ def record(FITS):
pack = ' : '.join(['Sex Parameters', key, value])
os.system('sethead %s HISTORY="%s"' %(FITS, pack))
return None
-
+
"""
if CATALOG_TYPE == 'FITS_LDAC': tbhdu = 2
@@ -662,7 +724,7 @@ def record(FITS):
#if MDIR is not None: record(FITS_SExCat)
AstSEx = Table.read(FITS_SExCat, hdu=tbhdu)
else: FITS_SExCat, AstSEx = None, None
-
+
PixA_SExCheckLst = []
FtmpLst = FITS_SExCheckLst.copy()
for k, FITS_SExCheck in enumerate(FITS_SExCheckLst):
@@ -691,15 +753,15 @@ def record(FITS):
MAGERR_TYPES = ['MAGERR_' + MAGT[4:] for MAGT in MAG_TYPES]
FLUX_TYPES = ['FLUX_' + MAGT[4:] for MAGT in MAG_TYPES]
FLUXERR_TYPES = ['FLUXERR_' + MAGT[4:] for MAGT in MAG_TYPES]
-
+
for i in range(len(MAG_TYPES)):
pcomplete = (MAGERR_TYPES[i] in USExParam) & \
(FLUX_TYPES[i] in USExParam) & \
(FLUXERR_TYPES[i] in USExParam)
-
+
if not pcomplete:
- _error_message = 'Please use complete FLUX FLUXERR MAG MAGERR '
+ _error_message = 'Please use complete FLUX FLUXERR MAG MAGERR '
_error_message += 'in SExParam for Negative Flux Correction!'
raise Exception('MeLOn ERROR [%s]: %s' %(objname, _error_message))
@@ -708,13 +770,13 @@ def record(FITS):
Mask_NC = FLUX < 0.0
MAG_NC = -2.5*np.log10(np.abs(FLUX[Mask_NC]))
MAGERR_NC = 1.0857 * np.abs(FLUXERR[Mask_NC] / FLUX[Mask_NC])
-
+
AstSEx[MAG_TYPES[i]][Mask_NC] = MAG_NC
AstSEx[MAGERR_TYPES[i]][Mask_NC] = MAGERR_NC
Modify_AstSEx = True
# ** b. ADD-COLUMN SEGLABEL if SEGMENTATION requested.
- # If some lines are discarded later, corresponding
+ # If some lines are discarded later, corresponding
# SEGLABEL will be lost in the output table.
if 'SEGMENTATION' in CHECKIMAGE_TYPE:
@@ -730,17 +792,17 @@ def record(FITS):
w_obj = Read_WCS.RW(phr_obj, VERBOSE_LEVEL=VERBOSE_LEVEL)
_XY = np.array([AstSEx['X_IMAGE'], AstSEx['Y_IMAGE']]).T
_RD = w_obj.all_pix2world(_XY, 1)
- AstSEx.add_column(Column(_RD[:, 0], name='X_WORLD'))
+ AstSEx.add_column(Column(_RD[:, 0], name='X_WORLD'))
AstSEx.add_column(Column(_RD[:, 1], name='Y_WORLD'))
if 'XWIN_IMAGE' in USExParam:
if 'YWIN_IMAGE' in USExParam:
_XY = np.array([AstSEx['XWIN_IMAGE'], AstSEx['YWIN_IMAGE']]).T
_RD = w_obj.all_pix2world(_XY, 1)
- AstSEx.add_column(Column(_RD[:, 0], name='XWIN_WORLD'))
+ AstSEx.add_column(Column(_RD[:, 0], name='XWIN_WORLD'))
AstSEx.add_column(Column(_RD[:, 1], name='YWIN_WORLD'))
Modify_AstSEx = True
-
+
# ** d. Restriction on FLAGS
if ONLY_FLAGS is not None:
if 'FLAGS' not in SExParam:
@@ -748,13 +810,13 @@ def record(FITS):
raise Exception('MeLOn ERROR [%s]: %s' %(objname, _error_message))
else:
_OLEN = len(AstSEx)
- AstSEx = AstSEx[np.in1d(AstSEx['FLAGS'], ONLY_FLAGS)]
+ AstSEx = AstSEx[np.in1d(AstSEx['FLAGS'], ONLY_FLAGS)]
Modify_AstSEx = True
if VERBOSE_LEVEL in [2]:
_message = 'PYSEx excludes [%d / %d] sources by FLAGS restriction!' %(_OLEN - len(AstSEx), _OLEN)
print('MeLOn CheckPoint [%s]: %s' %(objname, _message))
-
+
# ** e. Remove Boundary Sources
if XBoundary != 0.0 or XBoundary != 0.0:
NX, NY = int(phr_obj['NAXIS1']), int(phr_obj['NAXIS2'])
@@ -764,7 +826,7 @@ def record(FITS):
_XY[:, 0] < NX - XBoundary + 0.5, \
_XY[:, 1] > YBoundary + 0.5, \
_XY[:, 1] < NY - YBoundary + 0.5))
-
+
_OLEN = len(AstSEx)
AstSEx = AstSEx[InnerMask]
Modify_AstSEx = True
@@ -796,11 +858,11 @@ def record(FITS):
if RD_Quest is not None:
_RD = np.array([AstSEx[RAcoln_4Match], AstSEx[DECcoln_4Match]]).T
Symm = Sky_Symmetric_Match.SSM(RD_A=RD_Quest, RD_B=_RD, tol=Match_rdtol, return_distance=False)
-
+
if Symm is not None:
Modify_AstSEx = True
if Preserve_NoMatch:
- QuestMATCH = np.zeros(len(AstSEx)).astype(bool)
+ QuestMATCH = np.zeros(len(AstSEx)).astype(bool)
QuestMATCH[Symm[:, 1]] = True
AstSEx.add_column(Column(QuestMATCH, name='QuestMATCH'))
@@ -810,13 +872,13 @@ def record(FITS):
else:
_OLEN = len(AstSEx)
AstSEx = AstSEx[Symm[:, 1]]
-
+
QuestMATCH = np.ones(len(AstSEx)).astype(bool)
AstSEx.add_column(Column(QuestMATCH, name='QuestMATCH'))
-
+
QuestINDEX = Symm[:, 0]
AstSEx.add_column(Column(QuestINDEX, name='QuestINDEX'))
-
+
if VERBOSE_LEVEL in [2]:
_message = 'PYSEx excludes [%d / %d] sources by symmetric matching!' %(_OLEN - len(AstSEx), _OLEN)
print('MeLOn CheckPoint [%s]: %s' %(objname, _message))
@@ -833,7 +895,7 @@ def record(FITS):
FILL_VALUE=np.nan, FITS_StpLst=None, VERBOSE_LEVEL=VERBOSE_LEVEL)
AstSEx.add_column(Column(PixA_StpLst, name='Stamp'))
Modify_AstSEx = True
-
+
# ** UPDATE the file FITS_SExCat
if MDIR is not None and Modify_AstSEx:
tFITS_SExCat = ''.join([TDIR, '/TMPCAT_%s' %FNAME])
@@ -848,7 +910,7 @@ def record(FITS):
os.system('rm %s' %tFITS_SExCat)
# ** REMOVE temporary directory
- if MDIR is None:
+ if MDIR is None:
os.system('rm -rf %s' %TDIR)
return AstSEx, PixA_SExCheckLst, FITS_SExCat, FITS_SExCheckLst
From 4791ea08d8affe8d8630964fc079b57e7502de9d Mon Sep 17 00:00:00 2001
From: "Rob Knop (Nersc)" <raknop@lbl.gov>
Date: Mon, 6 Jan 2025 11:12:43 -0800
Subject: [PATCH 2/3] Add missing CudaResampling.pu
---
sfft/utils/CudaResampling.py | 410 +++++++++++++++++++++++++++++++++++
1 file changed, 410 insertions(+)
create mode 100644 sfft/utils/CudaResampling.py
diff --git a/sfft/utils/CudaResampling.py b/sfft/utils/CudaResampling.py
new file mode 100644
index 0000000..20a738b
--- /dev/null
+++ b/sfft/utils/CudaResampling.py
@@ -0,0 +1,410 @@
+import time
+import warnings
+import cupy as cp
+import numpy as np
+from math import floor
+from astropy.wcs import WCS, FITSFixedWarning
+from sfft.utils.CupyWCSTransform import Cupy_WCS_Transform
+
+__last_update__ = "2024-09-19"
+__author__ = "Lei Hu <leihu@andrew.cmu.edu>"
+
+class Cuda_Resampling:
+ def __init__(self, RESAMP_METHOD="BILINEAR", VERBOSE_LEVEL=2):
+ """
+ Image resampling using CUDA.
+ """
+ self.RESAMP_METHOD = RESAMP_METHOD
+ self.VERBOSE_LEVEL = VERBOSE_LEVEL
+ return None
+
+ def projection_cd(self, hdr_obj, hdr_targ, CDKEY="CD"):
+ """Mapping the target pixel centers to the object frame using Cupy for CD WCS (NO distortion)"""
+ NTX = int(hdr_targ["NAXIS1"])
+ NTY = int(hdr_targ["NAXIS2"])
+
+ XX_targ_GPU, YY_targ_GPU = cp.meshgrid(
+ cp.arange(0, NTX) + 1.,
+ cp.arange(0, NTY) + 1.,
+ indexing='ij'
+ )
+
+ CWT = Cupy_WCS_Transform()
+ # * perform CD transformation through target WCS
+ if True:
+ # read header [target]
+ KEYDICT, CD_GPU = CWT.read_cd_wcs(hdr_wcs=hdr_targ, CDKEY=CDKEY)
+ CRPIX1_targ, CRPIX2_targ = KEYDICT["CRPIX1"], KEYDICT["CRPIX2"]
+
+ # relative to reference point [target]
+ u_GPU, v_GPU = XX_targ_GPU.flatten() - CRPIX1_targ, YY_targ_GPU.flatten() - CRPIX2_targ
+
+ # CD transformation [target]
+ x_GPU, y_GPU = CWT.cd_transform(IMAGE_X_GPU=u_GPU, IMAGE_Y_GPU=v_GPU, CD_GPU=CD_GPU)
+
+ # * perform CD^-1 transformation through object WCS
+ if True:
+ # read header [object]
+ KEYDICT, CD_GPU = CWT.read_cd_wcs(hdr_wcs=hdr_obj, CDKEY=CDKEY)
+ CRPIX1_obj, CRPIX2_obj = KEYDICT["CRPIX1"], KEYDICT["CRPIX2"]
+
+ # relative to reference point, consider offset between the WCS reference points [object]
+ # WARNING: the offset calculations may be wrong when the WCS reference points are very close to the N/S Poles!
+ x_GPU += (CRPIX1_targ - CRPIX1_obj + 180.0) % 360.0 - 180.0 # from using target reference point to object
+ y_GPU += CRPIX2_targ - CRPIX2_obj # ~
+
+ # inverse CD transformation [object]
+ u_GPU, v_GPU = CWT.cd_transform_inv(WORLD_X_GPU=x_GPU, WORLD_Y_GPU=y_GPU, CD_GPU=CD_GPU)
+
+ # relative to image origin [object]
+ XX_proj_GPU = CRPIX1_obj + u_GPU.reshape((NTX, NTY))
+ YY_proj_GPU = CRPIX2_obj + v_GPU.reshape((NTX, NTY))
+
+ return XX_proj_GPU, YY_proj_GPU
+
+ def projection_sip(self, hdr_obj, hdr_targ, Nsamp=1024, RANDOM_SEED=10086):
+ """Mapping the target pixel centers to the object frame using Cupy for SIP WCS"""
+ NTX = int(hdr_targ["NAXIS1"])
+ NTY = int(hdr_targ["NAXIS2"])
+
+ XX_targ_GPU, YY_targ_GPU = cp.meshgrid(
+ cp.arange(0, NTX) + 1.,
+ cp.arange(0, NTY) + 1.,
+ indexing='ij'
+ )
+
+ CWT = Cupy_WCS_Transform()
+ # * perform forward transformation (+CD) through target WCS
+ if True:
+ # read header [target]
+ KEYDICT, CD_GPU, A_SIP_GPU, B_SIP_GPU = CWT.read_sip_wcs(hdr_wcs=hdr_targ)
+ CRPIX1_targ, CRPIX2_targ = KEYDICT["CRPIX1"], KEYDICT["CRPIX2"]
+ CRVAL1_targ, CRVAL2_targ = KEYDICT["CRVAL1"], KEYDICT["CRVAL2"]
+
+ # relative to reference point [target]
+ u_GPU, v_GPU = XX_targ_GPU.flatten() - CRPIX1_targ, YY_targ_GPU.flatten() - CRPIX2_targ
+
+ # forward transformation for target grid of pixel centers [target]
+ U_GPU, V_GPU = CWT.sip_forward_transform(u_GPU=u_GPU, v_GPU=v_GPU, A_SIP_GPU=A_SIP_GPU, B_SIP_GPU=B_SIP_GPU)
+
+ # plus CD transformation [target]
+ x_GPU, y_GPU = CWT.cd_transform(IMAGE_X_GPU=U_GPU, IMAGE_Y_GPU=V_GPU, CD_GPU=CD_GPU)
+
+ # * perform backward transformation (+CD^-1) through object WCS
+ if True:
+ # read header [object]
+ KEYDICT, CD_GPU, A_SIP_GPU, B_SIP_GPU = CWT.read_sip_wcs(hdr_wcs=hdr_obj)
+ CRPIX1_obj, CRPIX2_obj = KEYDICT["CRPIX1"], KEYDICT["CRPIX2"]
+ CRVAL1_obj, CRVAL2_obj = KEYDICT["CRVAL1"], KEYDICT["CRVAL2"]
+
+ # sampling random coordinates [object]
+ u_GPU, v_GPU = CWT.random_coord_sampling(N0=KEYDICT["N0"], N1=KEYDICT["N1"],
+ CRPIX1=CRPIX1_obj, CRPIX2=CRPIX2_obj, Nsamp=Nsamp, RANDOM_SEED=RANDOM_SEED)
+
+ # fit polynomial form backward transformation [object]
+ U_GPU, V_GPU = CWT.sip_forward_transform(u_GPU=u_GPU, v_GPU=v_GPU, A_SIP_GPU=A_SIP_GPU, B_SIP_GPU=B_SIP_GPU)
+ AP_lstsq_GPU, BP_lstsq_GPU = CWT.lstsq_sip_backward_transform(u_GPU=u_GPU, v_GPU=v_GPU,
+ U_GPU=U_GPU, V_GPU=V_GPU, A_ORDER=KEYDICT["A_ORDER"], B_ORDER=KEYDICT["B_ORDER"])[2:4]
+
+ # relative to reference point, consider offset between the WCS reference points [object]
+ # WARNING: the offset calculations may be wrong when the WCS reference points are very close to the N/S Poles!
+ # TODO: this is not accruate, ~ 1pix.
+ offset1 = ((CRVAL1_targ - CRVAL1_obj + 180.0) % 360.0 - 180.0) * cp.cos(cp.deg2rad((CRVAL2_targ + CRVAL2_obj)/2.))
+ offset2 = CRVAL2_targ - CRVAL2_obj
+ x_GPU += offset1
+ y_GPU += offset2
+
+ # inverse CD transformation [object]
+ U_GPU, V_GPU = CWT.cd_transform_inv(WORLD_X_GPU=x_GPU, WORLD_Y_GPU=y_GPU, CD_GPU=CD_GPU)
+
+ # backward transformation [object]
+ FP_UV_GPU = CWT.sip_backward_matrix(U_GPU=U_GPU, V_GPU=V_GPU, ORDER=KEYDICT["A_ORDER"])
+ GP_UV_GPU = CWT.sip_backward_matrix(U_GPU=U_GPU, V_GPU=V_GPU, ORDER=KEYDICT["B_ORDER"])
+
+ u_GPU, v_GPU = CWT.sip_backward_transform(U_GPU=U_GPU, V_GPU=V_GPU,
+ FP_UV_GPU=FP_UV_GPU, GP_UV_GPU=GP_UV_GPU, AP_GPU=AP_lstsq_GPU, BP_GPU=BP_lstsq_GPU)
+
+ # relative to image origin [object]
+ XX_proj_GPU = CRPIX1_obj + u_GPU.reshape((NTX, NTY))
+ YY_proj_GPU = CRPIX2_obj + v_GPU.reshape((NTX, NTY))
+
+ return XX_proj_GPU, YY_proj_GPU
+
+ def projection_astropy(self, hdr_obj, hdr_targ):
+ """Mapping the target pixel centers to the object frame using Astropy"""
+ # * read object WCS and target WCS
+ def _readWCS(hdr, VERBOSE_LEVEL=2):
+ with warnings.catch_warnings():
+ if VERBOSE_LEVEL in [0, 1]: behavior = 'ignore'
+ if VERBOSE_LEVEL in [2]: behavior = 'default'
+ warnings.filterwarnings(behavior, category=FITSFixedWarning)
+
+ if hdr['CTYPE1'] == 'RA---TAN' and 'PV1_0' in hdr:
+ _hdr = hdr.copy()
+ _hdr['CTYPE1'] = 'RA---TPV'
+ _hdr['CTYPE2'] = 'DEC--TPV'
+ else: _hdr = hdr
+ w = WCS(_hdr)
+ return w
+
+ w_obj = _readWCS(hdr=hdr_obj, VERBOSE_LEVEL=self.VERBOSE_LEVEL)
+ w_targ = _readWCS(hdr=hdr_targ, VERBOSE_LEVEL=self.VERBOSE_LEVEL)
+
+ # * maaping target pixel centers to the object frame
+ NTX = int(hdr_targ["NAXIS1"])
+ NTY = int(hdr_targ["NAXIS2"])
+
+ XX_targ, YY_targ = np.meshgrid(np.arange(0, NTX)+1., np.arange(0, NTY)+1., indexing='ij')
+ XY_targ = np.array([XX_targ.flatten(), YY_targ.flatten()]).T
+ XY_proj = w_obj.all_world2pix(w_targ.all_pix2world(XY_targ, 1), 1)
+
+ XX_proj = XY_proj[:, 0].reshape((NTX, NTY))
+ YY_proj = XY_proj[:, 1].reshape((NTX, NTY))
+
+ XX_proj_GPU = cp.array(XX_proj, dtype=np.float64)
+ YY_proj_GPU = cp.array(YY_proj, dtype=np.float64)
+
+ return XX_proj_GPU, YY_proj_GPU
+
+ def frame_extension(self, XX_proj_GPU, YY_proj_GPU, PixA_obj_GPU):
+ """Extend the object frame for resampling"""
+ NTX, NTY = XX_proj_GPU.shape
+ NOX, NOY = PixA_obj_GPU.shape
+
+ # * padding the object frame
+ if self.RESAMP_METHOD == 'BILINEAR':
+ KERHW = (1, 1)
+
+ if self.RESAMP_METHOD == 'LANCZOS3':
+ KERHW = (3, 3)
+
+ # find the root index and shift with maximal kernel halfwidth (KERHW)
+ # Note: the index ranges (RMIN --- RMAX) and (CMIN --- CMAX) can cover
+ # all pixels that interpolation may be use.
+
+ RMIN = (floor(XX_proj_GPU.min().item()) - 1) - KERHW[0]
+ RMAX = (floor(XX_proj_GPU.max().item()) - 1) + KERHW[0]
+ RPAD = (-np.min([RMIN, 0]), np.max([RMAX - (NOX - 1), 0]))
+
+ CMIN = (floor(YY_proj_GPU.min().item()) - 1) - KERHW[1]
+ CMAX = (floor(YY_proj_GPU.max().item()) - 1) + KERHW[1]
+ CPAD = (-np.min([CMIN, 0]), np.max([CMAX - (NOY - 1), 0]))
+
+ PAD_WIDTH = (RPAD, CPAD)
+ PixA_Eobj_GPU = cp.pad(PixA_obj_GPU, PAD_WIDTH, mode='constant', constant_values=0.)
+ NEOX, NEOY = PixA_Eobj_GPU.shape
+
+ XX_Eproj_GPU = XX_proj_GPU + PAD_WIDTH[0][0]
+ YY_Eproj_GPU = YY_proj_GPU + PAD_WIDTH[1][0]
+
+ RMIN_E = (floor(XX_Eproj_GPU.min().item()) - 1) - KERHW[0]
+ RMAX_E = (floor(XX_Eproj_GPU.max().item()) - 1) + KERHW[0]
+
+ CMIN_E = (floor(YY_Eproj_GPU.min().item()) - 1) - KERHW[1]
+ CMAX_E = (floor(YY_Eproj_GPU.max().item()) - 1) + KERHW[1]
+
+ assert RMIN_E >= 0 and CMIN_E >= 0
+ assert RMAX_E < NEOX and CMAX_E < NEOY
+
+ EProjDict = {}
+ EProjDict['NTX'] = NTX
+ EProjDict['NTY'] = NTY
+
+ EProjDict['NOX'] = NOX
+ EProjDict['NOY'] = NOY
+
+ EProjDict['NEOX'] = NEOX
+ EProjDict['NEOY'] = NEOY
+
+ EProjDict['XX_Eproj_GPU'] = XX_Eproj_GPU
+ EProjDict['YY_Eproj_GPU'] = YY_Eproj_GPU
+
+ return PixA_Eobj_GPU, EProjDict
+
+ def resampling(self, PixA_Eobj_GPU, EProjDict):
+ """Resampling the object frame to the target frame using CUDA"""
+ NTX = EProjDict['NTX']
+ NTY = EProjDict['NTY']
+
+ NEOX = EProjDict['NEOX']
+ NEOY = EProjDict['NEOY']
+
+ XX_Eproj_GPU = EProjDict['XX_Eproj_GPU']
+ YY_Eproj_GPU = EProjDict['YY_Eproj_GPU']
+
+ # * Cupy configuration
+ MaxThreadPerB = 8
+ GPUManage = lambda NT: ((NT-1)//MaxThreadPerB + 1, min(NT, MaxThreadPerB))
+ BpG_PIX0, TpB_PIX0 = GPUManage(NTX)
+ BpG_PIX1, TpB_PIX1 = GPUManage(NTY)
+ BpG_PIX, TpB_PIX = (BpG_PIX0, BpG_PIX1), (TpB_PIX0, TpB_PIX1, 1)
+
+ PixA_resamp_GPU = cp.zeros((NTX, NTY), dtype=np.float64)
+
+ if self.RESAMP_METHOD == "BILINEAR":
+
+ # * perform bilinear resampling using CUDA
+ # input: PixA_Eobj | (NEOX, NEOY)
+ # input: XX_Eproj, YY_Eproj | (NTX, NTY)
+ # output: PixA_resamp | (NTX, NTY)
+
+ _refdict = {'NTX': NTX, 'NTY': NTY, 'NEOX': NEOX, 'NEOY': NEOY}
+ _funcstr = r"""
+ extern "C" __global__ void kmain(double XX_Eproj_GPU[%(NTX)s][%(NTY)s], double YY_Eproj_GPU[%(NTX)s][%(NTY)s],
+ double PixA_Eobj_GPU[%(NEOX)s][%(NEOY)s], double PixA_resamp_GPU[%(NTX)s][%(NTY)s])
+ {
+ int ROW = blockIdx.x*blockDim.x+threadIdx.x;
+ int COL = blockIdx.y*blockDim.y+threadIdx.y;
+
+ int NTX = %(NTX)s;
+ int NTY = %(NTY)s;
+ int NEOX = %(NEOX)s;
+ int NEOY = %(NEOY)s;
+
+ if (ROW < NTX && COL < NTY) {
+
+ double x = XX_Eproj_GPU[ROW][COL];
+ double y = YY_Eproj_GPU[ROW][COL];
+
+ int r1 = floor(x) - 1;
+ int c1 = floor(y) - 1;
+ int r2 = r1 + 1;
+ int c2 = c1 + 1;
+
+ double dx = x - floor(x);
+ double dy = y - floor(y);
+
+ double w11 = (1-dx) * (1-dy);
+ double w12 = (1-dx) * dy;
+ double w21 = dx * (1-dy);
+ double w22 = dx * dy;
+
+ PixA_resamp_GPU[ROW][COL] = w11 * PixA_Eobj_GPU[r1][c1] + w12 * PixA_Eobj_GPU[r1][c2] +
+ w21 * PixA_Eobj_GPU[r2][c1] + w22 * PixA_Eobj_GPU[r2][c2];
+ }
+ }
+ """
+ _code = _funcstr % _refdict
+ _module = cp.RawModule(code=_code, backend=u'nvcc', translate_cucomplex=False)
+ resamp_func = _module.get_function('kmain')
+
+ t0 = time.time()
+ resamp_func(args=(XX_Eproj_GPU, YY_Eproj_GPU, PixA_Eobj_GPU, PixA_resamp_GPU), block=TpB_PIX, grid=BpG_PIX)
+ if self.VERBOSE_LEVEL in [1, 2]:
+ print('MeLOn CheckPoint: Cuda resampling takes [%.6f s]' %(time.time() - t0))
+
+ if self.RESAMP_METHOD == "LANCZOS3":
+
+ # * perform LANCZOS-3 resampling using CUDA
+ # input: XX_Eproj, YY_Eproj | (NTX, NTY)
+ # output: PixA_resamp | (NTX, NTY)
+
+ _refdict = {'NTX': NTX, 'NTY': NTY}
+ _funcstr = r"""
+ extern "C" __global__ void kmain(double XX_Eproj_GPU[%(NTX)s][%(NTY)s], double YY_Eproj_GPU[%(NTX)s][%(NTY)s],
+ double LKERNEL_X_GPU[6][%(NTX)s][%(NTY)s], double LKERNEL_Y_GPU[6][%(NTX)s][%(NTY)s])
+ {
+ int ROW = blockIdx.x*blockDim.x+threadIdx.x;
+ int COL = blockIdx.y*blockDim.y+threadIdx.y;
+
+ int NTX = %(NTX)s;
+ int NTY = %(NTY)s;
+
+ double PI = 3.141592653589793;
+ double PIS = 9.869604401089358;
+
+ if (ROW < NTX && COL < NTY) {
+
+ double x = XX_Eproj_GPU[ROW][COL];
+ double y = YY_Eproj_GPU[ROW][COL];
+
+ double dx = x - floor(x);
+ double dy = y - floor(y);
+
+ // LANCZOS3 weights in x axis
+ double wx0 = 3.0 * sin(PI*(-2.0 - dx)) * sin(PI*(-2.0 - dx)/3.0) / (PIS*(-2.0 - dx) * (-2.0 - dx));
+ double wx1 = 3.0 * sin(PI*(-1.0 - dx)) * sin(PI*(-1.0 - dx)/3.0) / (PIS*(-1.0 - dx) * (-1.0 - dx));
+ double wx2 = 1.0;
+ if (fabs(dx) > 1e-4) {
+ wx2 = 3.0 * sin(PI*(-dx)) * sin(PI*(-dx)/3.0) / (PIS*(-dx) * (-dx));
+ }
+ double wx3 = 3.0 * sin(PI*(1.0 - dx)) * sin(PI*(1.0 - dx)/3.0) / (PIS*(1.0 - dx) * (1.0 - dx));
+ double wx4 = 3.0 * sin(PI*(2.0 - dx)) * sin(PI*(2.0 - dx)/3.0) / (PIS*(2.0 - dx) * (2.0 - dx));
+ double wx5 = 3.0 * sin(PI*(3.0 - dx)) * sin(PI*(3.0 - dx)/3.0) / (PIS*(3.0 - dx) * (3.0 - dx));
+
+ LKERNEL_X_GPU[0][ROW][COL] = wx0;
+ LKERNEL_X_GPU[1][ROW][COL] = wx1;
+ LKERNEL_X_GPU[2][ROW][COL] = wx2;
+ LKERNEL_X_GPU[3][ROW][COL] = wx3;
+ LKERNEL_X_GPU[4][ROW][COL] = wx4;
+ LKERNEL_X_GPU[5][ROW][COL] = wx5;
+
+ // LANCZOS3 weights in y axis
+ double wy0 = 3.0 * sin(PI*(-2.0 - dy)) * sin(PI*(-2.0 - dy)/3.0) / (PIS*(-2.0 - dy) * (-2.0 - dy));
+ double wy1 = 3.0 * sin(PI*(-1.0 - dy)) * sin(PI*(-1.0 - dy)/3.0) / (PIS*(-1.0 - dy) * (-1.0 - dy));
+ double wy2 = 1.0;
+ if (fabs(dy) > 1e-4) {
+ wy2 = 3.0 * sin(PI*(-dy)) * sin(PI*(-dy)/3.0) / (PIS*(-dy) * (-dy));
+ }
+ double wy3 = 3.0 * sin(PI*(1.0 - dy)) * sin(PI*(1.0 - dy)/3.0) / (PIS*(1.0 - dy) * (1.0 - dy));
+ double wy4 = 3.0 * sin(PI*(2.0 - dy)) * sin(PI*(2.0 - dy)/3.0) / (PIS*(2.0 - dy) * (2.0 - dy));
+ double wy5 = 3.0 * sin(PI*(3.0 - dy)) * sin(PI*(3.0 - dy)/3.0) / (PIS*(3.0 - dy) * (3.0 - dy));
+
+ LKERNEL_Y_GPU[0][ROW][COL] = wy0;
+ LKERNEL_Y_GPU[1][ROW][COL] = wy1;
+ LKERNEL_Y_GPU[2][ROW][COL] = wy2;
+ LKERNEL_Y_GPU[3][ROW][COL] = wy3;
+ LKERNEL_Y_GPU[4][ROW][COL] = wy4;
+ LKERNEL_Y_GPU[5][ROW][COL] = wy5;
+ }
+ }
+ """
+ _code = _funcstr % _refdict
+ _module = cp.RawModule(code=_code, backend=u'nvcc', translate_cucomplex=False)
+ weightkernel_func = _module.get_function('kmain')
+
+ _refdict = {'NTX': NTX, 'NTY': NTY, 'NEOX': NEOX, 'NEOY': NEOY}
+ _funcstr = r"""
+ extern "C" __global__ void kmain(double XX_Eproj_GPU[%(NTX)s][%(NTY)s], double YY_Eproj_GPU[%(NTX)s][%(NTY)s],
+ double LKERNEL_X_GPU[6][%(NTX)s][%(NTY)s], double LKERNEL_Y_GPU[6][%(NTX)s][%(NTY)s],
+ double PixA_Eobj_GPU[%(NEOX)s][%(NEOY)s], double PixA_resamp_GPU[%(NTX)s][%(NTY)s])
+ {
+ int ROW = blockIdx.x*blockDim.x+threadIdx.x;
+ int COL = blockIdx.y*blockDim.y+threadIdx.y;
+
+ int NTX = %(NTX)s;
+ int NTY = %(NTY)s;
+ int NEOX = %(NEOX)s;
+ int NEOY = %(NEOY)s;
+
+ if (ROW < NTX && COL < NTY) {
+
+ double x = XX_Eproj_GPU[ROW][COL];
+ double y = YY_Eproj_GPU[ROW][COL];
+
+ int r0 = floor(x) - 3;
+ int c0 = floor(y) - 3;
+
+ for(int i = 0; i < 6; ++i){
+ for(int j = 0; j < 6; ++j){
+ double w = LKERNEL_X_GPU[i][ROW][COL] * LKERNEL_Y_GPU[j][ROW][COL];
+ PixA_resamp_GPU[ROW][COL] += w * PixA_Eobj_GPU[r0 + i][c0 + j];
+ }
+ }
+ }
+ }
+ """
+ _code = _funcstr % _refdict
+ _module = cp.RawModule(code=_code, backend=u'nvcc', translate_cucomplex=False)
+ resamp_func = _module.get_function('kmain')
+
+ t0 = time.time()
+ LKERNEL_X_GPU = cp.zeros((6, NTX, NTY), dtype=np.float64)
+ LKERNEL_Y_GPU = cp.zeros((6, NTX, NTY), dtype=np.float64)
+ weightkernel_func(args=(XX_Eproj_GPU, YY_Eproj_GPU, LKERNEL_X_GPU, LKERNEL_Y_GPU), block=TpB_PIX, grid=BpG_PIX)
+ resamp_func(args=(XX_Eproj_GPU, YY_Eproj_GPU, LKERNEL_X_GPU, LKERNEL_Y_GPU,
+ PixA_Eobj_GPU, PixA_resamp_GPU), block=TpB_PIX, grid=BpG_PIX)
+ if self.VERBOSE_LEVEL in [1, 2]:
+ print('MeLOn CheckPoint: Cuda resampling takes [%.6f s]' %(time.time() - t0))
+
+ return PixA_resamp_GPU
From eee96453cd40e12e62c5251d69fd43ae533a8c95 Mon Sep 17 00:00:00 2001
From: "Rob Knop (Nersc)" <raknop@lbl.gov>
Date: Tue, 7 Jan 2025 10:05:58 -0800
Subject: [PATCH 3/3] Add missing CupyWCSTransform.py
---
sfft/utils/CupyWCSTransform.py | 241 +++++++++++++++++++++++++++++++++
1 file changed, 241 insertions(+)
create mode 100644 sfft/utils/CupyWCSTransform.py
diff --git a/sfft/utils/CupyWCSTransform.py b/sfft/utils/CupyWCSTransform.py
new file mode 100644
index 0000000..ceb47b0
--- /dev/null
+++ b/sfft/utils/CupyWCSTransform.py
@@ -0,0 +1,241 @@
+import cupy as cp
+
+__last_update__ = "2024-09-19"
+__author__ = "Shu Liu <shl159@pitt.edu> & Lei Hu <leihu@andrew.cmu.edu>"
+
+class Cupy_WCS_Transform:
+ def __init__(self):
+ return None
+
+ def read_cd_wcs(self, hdr_wcs, CDKEY="CD"):
+ """
+ # * Note on the CD matrix transformation:
+ The sky coordinate (x, y) relative to reference point can be connected with
+ the image coordinate (u, v) relative to reference point by the CD matrix:
+ [x] [CD1_1 CD1_2] [u]
+ [y] = [CD2_1 CD2_2] [v]
+ where CD1_1, CD1_2, CD2_1, CD2_2 are stored in the FITS header.
+
+ """
+ assert hdr_wcs["CTYPE1"] == "RA---TAN"
+ assert hdr_wcs["CTYPE2"] == "DEC--TAN"
+
+ N0 = int(hdr_wcs["NAXIS1"])
+ N1 = int(hdr_wcs["NAXIS2"])
+
+ CRPIX1 = float(hdr_wcs["CRPIX1"])
+ CRPIX2 = float(hdr_wcs["CRPIX2"])
+
+ CRVAL1 = float(hdr_wcs["CRVAL1"])
+ CRVAL2 = float(hdr_wcs["CRVAL2"])
+
+ KEYDICT = {
+ "N0": N0, "N1": N1,
+ "CRPIX1": CRPIX1, "CRPIX2": CRPIX2,
+ "CRVAL1": CRVAL1, "CRVAL2": CRVAL2
+ }
+
+ CD1_1 = hdr_wcs[f"{CDKEY}1_1"]
+ CD1_2 = hdr_wcs[f"{CDKEY}1_2"] if f"{CDKEY}1_2" in hdr_wcs else 0.
+ CD2_1 = hdr_wcs[f"{CDKEY}2_1"] if f"{CDKEY}2_1" in hdr_wcs else 0.
+ CD2_2 = hdr_wcs[f"{CDKEY}2_2"]
+
+ CD_GPU = cp.array([
+ [CD1_1, CD1_2],
+ [CD2_1, CD2_2]
+ ], dtype=cp.float64)
+
+ return KEYDICT, CD_GPU
+
+ def read_sip_wcs(self, hdr_wcs):
+ """
+ # * Note on the SIP transformation:
+ The image coordinate (u, v) relative to reference point and we have undistorted image coordinate (U, V):
+ U = u + f(u, v)
+ V = v + g(u, v)
+ where f(u, v) and g(u, v) are the SIP distortion functions with polynomial form:
+ f(u, v) = sum_{p, q} A_{pq} u^p v^q, p + q <= A_ORDER
+ g(u, v) = sum_{p, q} B_{pq} u^p v^q, p + q <= B_ORDER
+ These coefficients are stored in the FITS header and define a foward transformation from (u, v) to (U, V).
+
+ The sky coordinate (x, y) relative to reference point can be connected with (U, V) by the reversible CD matrix:
+ [x] [CD1_1 CD1_2] [U]
+ [y] = [CD2_1 CD2_2] [V]
+ So the forward transformation & CD matrix (with considering the reference point) is equivalent to a pix2world function.
+
+ The reverse question is how to obtain the backward transformation from undistorted (U, V) to distorted (u, v).
+ Once we have the backward transformation, we can combine a reversed CD matrix
+ (with considering the reference point) to get the world2pix function.
+
+ A simple approach (not accurate) is to assume the forward transformation also has a polynomial form:
+ u = U + fp(U, V)
+ v = V + gp(U, V)
+ where
+ fp(U, V) = sum_{p, q} AP_{pq} U^p V^q, p + q <= A_ORDER
+ gp(U, V) = sum_{p, q} BP_{pq} U^p V^q, p + q <= B_ORDER
+
+ The coefficients AP_{pq} and BP_{pq} can be obtained by solving the linear equations separately.
+ [u - U] = [U^p*V^q] [AP_{pq}]
+
+ shapes:
+ b: [u - U] | (N, 1),
+ A: [U^p*V^q] | (N, (A_ORDER+1)*(A_ORDER+2)/2)
+ x: [AP_{pq}] | ((A_ORDER+1)*(A_ORDER+2)/2, 1)
+
+ [v - V] = [U^p*V^q] [BP_{pq}]
+ b: [v - V] | (N, 1),
+ A: [U^p*V^q] | (N, (B_ORDER+1)*(B_ORDER+2)/2)
+ x: [BP_{pq}] | ((B_ORDER+1)*(B_ORDER+2)/2, 1)
+
+ The two matrices encode the "backward" transformation in fp(U, V) and gp(U, V).
+ Let's denote the two matrices as FP_UV and GP_UV, respectively.
+
+ # * A WCS example with SIP distortion
+ CTYPE1 = 'RA---TAN-SIP'
+ CTYPE2 = 'DEC--TAN-SIP'
+
+ CRPIX1 = 2044.0
+ CRPIX2 = 2044.0
+
+ CD1_1 = 2.65458074767927E-05
+ CD1_2 = -1.2630331175158E-05
+ CD2_1 = 1.38917634264203E-05
+ CD2_2 = 2.57785917553667E-05
+
+ CRVAL1 = 9.299707734492053
+ CRVAL2 = -43.984663860136145
+
+ A_ORDER = 4
+ A_0_2 = -4.774423702E-10
+ A_0_3 = -2.462744787E-14
+ A_0_4 = 2.28913156E-17
+ A_1_1 = 1.076615801E-09
+ A_1_2 = -9.939904553E-14
+ A_1_3 = -5.845336863E-17
+ A_2_0 = -3.717865118E-10
+ A_2_1 = -4.966118494E-15
+ A_2_2 = 6.615859199E-17
+ A_3_0 = -3.817793356E-15
+ A_3_1 = 3.310020049E-17
+ A_4_0 = -5.166048303E-19
+
+ B_ORDER = 4
+ B_0_2 = 1.504792792E-09
+ B_0_3 = -2.662833066E-15
+ B_0_4 = -8.383877471E-20
+ B_1_1 = 1.204553316E-10
+ B_1_2 = 6.57748238E-14
+ B_1_3 = -6.998508554E-18
+ B_2_0 = 4.136013664E-10
+ B_2_1 = 5.339208582E-14
+ B_2_2 = 6.063403412E-17
+ B_3_0 = 1.486316541E-14
+ B_3_1 = -9.102668806E-17
+ B_4_0 = -5.174323631E-17
+
+ """
+ assert hdr_wcs["CTYPE1"] == "RA---TAN-SIP"
+ assert hdr_wcs["CTYPE2"] == "DEC--TAN-SIP"
+
+ N0 = int(hdr_wcs["NAXIS1"])
+ N1 = int(hdr_wcs["NAXIS2"])
+
+ CRPIX1 = float(hdr_wcs["CRPIX1"])
+ CRPIX2 = float(hdr_wcs["CRPIX2"])
+
+ CRVAL1 = float(hdr_wcs["CRVAL1"])
+ CRVAL2 = float(hdr_wcs["CRVAL2"])
+
+ A_ORDER = int(hdr_wcs["A_ORDER"])
+ B_ORDER = int(hdr_wcs["B_ORDER"])
+
+ KEYDICT = {
+ "N0": N0, "N1": N1,
+ "CRPIX1": CRPIX1, "CRPIX2": CRPIX2,
+ "CRVAL1": CRVAL1, "CRVAL2": CRVAL2,
+ "A_ORDER": A_ORDER, "B_ORDER": B_ORDER
+ }
+
+ CD_GPU = cp.array([
+ [hdr_wcs["CD1_1"], hdr_wcs["CD1_2"]],
+ [hdr_wcs["CD2_1"], hdr_wcs["CD2_2"]]
+ ], dtype=cp.float64)
+
+ A_SIP_GPU = cp.zeros((A_ORDER+1, A_ORDER+1), dtype=cp.float64)
+ B_SIP_GPU = cp.zeros((B_ORDER+1, B_ORDER+1), dtype=cp.float64)
+
+ for p in range(A_ORDER + 1):
+ for q in range(0, A_ORDER - p + 1):
+ keyword = f"A_{p}_{q}"
+ if keyword in hdr_wcs:
+ A_SIP_GPU[p, q] = hdr_wcs[keyword]
+
+ for p in range(B_ORDER + 1):
+ for q in range(0, B_ORDER - p + 1):
+ keyword = f"B_{p}_{q}"
+ if keyword in hdr_wcs:
+ B_SIP_GPU[p, q] = hdr_wcs[keyword]
+
+ return KEYDICT, CD_GPU, A_SIP_GPU, B_SIP_GPU
+
+ def cd_transform(self, IMAGE_X_GPU, IMAGE_Y_GPU, CD_GPU):
+ """CD matrix transformation from image to world"""
+ WORLD_X_GPU, WORLD_Y_GPU = cp.matmul(CD_GPU, cp.array([IMAGE_X_GPU, IMAGE_Y_GPU]))
+ return WORLD_X_GPU, WORLD_Y_GPU
+
+ def cd_transform_inv(self, WORLD_X_GPU, WORLD_Y_GPU, CD_GPU):
+ """CD matrix transformation from world to image"""
+ # CD_inv_GPU = cp.linalg.inv(CD_GPU)
+ CD_inv_GPU = 1. / (CD_GPU[0, 0] * CD_GPU[1, 1] - CD_GPU[0, 1] * CD_GPU[1, 0]) * \
+ cp.array([[CD_GPU[1, 1], -CD_GPU[0, 1]], [-CD_GPU[1, 0], CD_GPU[0, 0]]])
+ IMAGE_X_GPU, IMAGE_Y_GPU = cp.matmul(CD_inv_GPU, cp.array([WORLD_X_GPU, WORLD_Y_GPU]))
+ return IMAGE_X_GPU, IMAGE_Y_GPU
+
+ def sip_forward_transform(self, u_GPU, v_GPU, A_SIP_GPU, B_SIP_GPU):
+ """Forward SIP transformation from (u, v) to (U, V)"""
+ A_ORDER = A_SIP_GPU.shape[0] - 1
+ U_GPU = u_GPU + cp.zeros_like(u_GPU)
+ for p in range(A_ORDER + 1):
+ for q in range(A_ORDER - p + 1):
+ U_GPU += A_SIP_GPU[p, q] * u_GPU**p * v_GPU**q
+
+ B_ORDER = B_SIP_GPU.shape[0] - 1
+ V_GPU = v_GPU + cp.zeros_like(v_GPU)
+ for p in range(B_ORDER + 1):
+ for q in range(B_ORDER - p + 1):
+ V_GPU += B_SIP_GPU[p, q] * u_GPU**p * v_GPU**q
+ return U_GPU, V_GPU
+
+ def sip_backward_matrix(self, U_GPU, V_GPU, ORDER):
+ """Compute the backward matrix P_UV (FP_UV or GP_UV)"""
+ Nsamp = len(U_GPU)
+ P_UV_GPU = cp.zeros((Nsamp, (ORDER+1)*(ORDER+2)//2), dtype=cp.float64)
+ idx = 0
+ for p in range(ORDER + 1):
+ for q in range(ORDER - p + 1):
+ P_UV_GPU[:, idx] = U_GPU**p * V_GPU**q
+ idx += 1
+ return P_UV_GPU
+
+ def lstsq_sip_backward_transform(self, u_GPU, v_GPU, U_GPU, V_GPU, A_ORDER, B_ORDER):
+ """Compute the backward transformation from (U, V) to (u, v)"""
+ FP_UV_GPU = self.sip_backward_matrix(U_GPU=U_GPU, V_GPU=V_GPU, ORDER=A_ORDER)
+ GP_UV_GPU = self.sip_backward_matrix(U_GPU=U_GPU, V_GPU=V_GPU, ORDER=B_ORDER)
+
+ AP_lstsq_GPU = cp.linalg.lstsq(FP_UV_GPU, (u_GPU - U_GPU).reshape(-1, 1), rcond=None)[0]
+ BP_lstsq_GPU = cp.linalg.lstsq(GP_UV_GPU, (v_GPU - V_GPU).reshape(-1, 1), rcond=None)[0]
+ return FP_UV_GPU, GP_UV_GPU, AP_lstsq_GPU, BP_lstsq_GPU
+
+ def sip_backward_transform(self, U_GPU, V_GPU, FP_UV_GPU, GP_UV_GPU, AP_GPU, BP_GPU):
+ """Backward transformation from (U, V) to (u, v)"""
+ u_GPU = U_GPU + cp.matmul(FP_UV_GPU, AP_GPU)[:, 0]
+ v_GPU = V_GPU + cp.matmul(GP_UV_GPU, BP_GPU)[:, 0]
+ return u_GPU, v_GPU
+
+ def random_coord_sampling(self, N0, N1, CRPIX1, CRPIX2, Nsamp=1024, RANDOM_SEED=10086):
+ """Random sampling of coordinates"""
+ if RANDOM_SEED is not None:
+ cp.random.seed(RANDOM_SEED)
+ u_GPU = cp.random.uniform(0.5, N0+0.5, Nsamp, dtype=cp.float64) - CRPIX1
+ v_GPU = cp.random.uniform(0.5, N1+0.5, Nsamp, dtype=cp.float64) - CRPIX2
+ return u_GPU, v_GPU