lsst.ip.diffim gfaa7bcd731+3e3b9e0885
Loading...
Searching...
No Matches
KernelPca.cc
Go to the documentation of this file.
1// -*- lsst-c++ -*-
12#include "lsst/afw/math.h"
13#include "lsst/afw/image.h"
14#include "lsst/log/Log.h"
16
19
20namespace afwMath = lsst::afw::math;
21namespace afwImage = lsst::afw::image;
23
24namespace lsst {
25namespace ip {
26namespace diffim {
27namespace detail {
28
55 template<typename PixelT>
58 ) :
59 afwMath::CandidateVisitor(),
60 _imagePca(imagePca),
61 _mean()
62 {};
63
64 template<typename PixelT>
66 afwMath::KernelList kernelList;
67
68 std::vector<std::shared_ptr<ImageT>> eigenImages = _imagePca->getEigenImages();
69 int ncomp = eigenImages.size();
70
71 if (_mean) {
75 (*_mean, true))));
76 }
77 for (int i = 0; i < ncomp; i++) {
79 afwImage::Image<afwMath::Kernel::Pixel>(*eigenImages[i], true);
82 ));
83 }
84
85 return kernelList;
86 }
87
88 template<typename PixelT>
90
91 KernelCandidate<PixelT> *kCandidate = dynamic_cast<KernelCandidate<PixelT> *>(candidate);
92 if (kCandidate == NULL) {
94 "Failed to cast SpatialCellCandidate to KernelCandidate");
95 }
96 LOGL_DEBUG("TRACE5.ip.diffim.SetPcaImageVisitor.processCandidate",
97 "Processing candidate %d", kCandidate->getId());
98
99 try {
100 /* Normalize to unit sum */
101 std::shared_ptr<ImageT> kImage = kCandidate->getKernelSolution(
102 KernelCandidate<PixelT>::ORIG)->makeKernelImage();
103 *kImage /= kCandidate->getKernelSolution(
105 /* Tell imagePca they have the same weighting in the Pca */
106 _imagePca->addImage(kImage, 1.0);
107 } catch(pexExcept::Exception &e) {
108 return;
109 }
110 }
111
112 template<typename PixelT>
114 /*
115 If we don't subtract off the mean before we do the Pca, the
116 subsequent terms carry less of the power than if you do subtract
117 off the mean. Explicit example:
118
119 With mean subtraction:
120 DEBUG: Eigenvalue 0 : 0.010953 (0.373870 %)
121 DEBUG: Eigenvalue 1 : 0.007927 (0.270604 %)
122 DEBUG: Eigenvalue 2 : 0.001393 (0.047542 %)
123 DEBUG: Eigenvalue 3 : 0.001092 (0.037261 %)
124 DEBUG: Eigenvalue 4 : 0.000829 (0.028283 %)
125
126 Without mean subtraction:
127 DEBUG: Eigenvalue 0 : 0.168627 (0.876046 %)
128 DEBUG: Eigenvalue 1 : 0.007935 (0.041223 %)
129 DEBUG: Eigenvalue 2 : 0.006049 (0.031424 %)
130 DEBUG: Eigenvalue 3 : 0.001188 (0.006173 %)
131 DEBUG: Eigenvalue 4 : 0.001050 (0.005452 %)
132
133 After the first term above, which basically represents the mean,
134 the remaining terms carry less of the power than if you do
135 subtract off the mean. (0.041223/(1-0.876046) < 0.373870).
136 */
137 LOGL_DEBUG("TRACE5.ip.diffim.KernelPcaVisitor.subtractMean",
138 "Subtracting mean feature before Pca");
139
140 _mean = _imagePca->getMean();
141 KernelPca<ImageT>::ImageList imageList = _imagePca->getImageList();
142 for (typename KernelPca<ImageT>::ImageList::const_iterator ptr = imageList.begin(),
143 end = imageList.end(); ptr != end; ++ptr) {
144 **ptr -= *_mean;
145 }
146 }
147
162 template <typename ImageT>
164 {
165 Super::analyze();
166
167 typename Super::ImageList const &eImageList = this->getEigenImages();
168 typename Super::ImageList::const_iterator iter = eImageList.begin(), end = eImageList.end();
169 for (size_t i = 0; iter != end; ++i, ++iter) {
171
172 /*
173 * Normalise eigenImages to have a maximum of 1.0. For n > 0 they
174 * (should) have mean == 0, so we can't use that to normalize
175 */
177 double const min = stats.getValue(afwMath::MIN);
178 double const max = stats.getValue(afwMath::MAX);
179
180 double const extreme = (fabs(min) > max) ? min :max;
181 if (extreme != 0.0) {
182 *eImage /= extreme;
183 }
184 }
185 }
186
187
188 typedef float PixelT;
189 template class KernelPcaVisitor<PixelT>;
191
192}}}} // end of namespace lsst::ip::diffim::detail
#define LOGL_DEBUG(logger, message...)
int min
int end
int max
#define LSST_EXCEPT(type,...)
Class used by SpatialModelCell for spatial Kernel fitting.
Declaration of KernelPca and KernelPcaVisitor.
T begin(T... args)
double getValue(Property const prop=NOTHING) const
Class stored in SpatialCells for spatial Kernel fitting.
std::shared_ptr< StaticKernelSolution< PixelT > > getKernelSolution(CandidateSwitch cand) const
Overrides the analyze method of base class afwImage::ImagePca.
Definition: KernelPca.h:24
virtual void analyze()
Generate eigenimages that are normalised.
Definition: KernelPca.cc:163
A class to run a PCA on all candidate kernels (represented as Images).
Definition: KernelPca.h:40
void processCandidate(lsst::afw::math::SpatialCellCandidate *candidate)
Definition: KernelPca.cc:89
lsst::afw::math::KernelList getEigenKernels()
Definition: KernelPca.cc:65
KernelPcaVisitor(std::shared_ptr< KernelPca< ImageT > > imagePca)
Definition: KernelPca.cc:56
T end(T... args)
def iter(self)
Statistics makeStatistics(lsst::afw::image::Image< Pixel > const &img, lsst::afw::image::Mask< image::MaskPixel > const &msk, int const flags, StatisticsControl const &sctrl=StatisticsControl())
T push_back(T... args)
T size(T... args)