IPP Software Navigation Tools IPP Links Communication Pan-STARRS Links

Ignore:
Timestamp:
Aug 31, 2013, 5:55:16 AM (13 years ago)
Author:
eugene
Message:

merge changes from EAM branch (pmModel_CentralPixel functions for sersic-like model support; add interactive mode to PCM LMM fitting; harder PCM LMM damping (nu scaled by 3 not 2); add EXT_AND SKY and SHAPE fitting modes, include sky in INDEX-only fits; ifdefs for trace-only values; allow positions to go to 100,000; major re-work of central pixel & flux for sersic-like models; add pmPCMMakeModel function to generate the flux for an arbitrary model

Location:
trunk/psModules
Files:
2 edited

Legend:

Unmodified
Added
Removed

  • trunk/psModules

  • trunk/psModules/src/objects/models/pmModel_SERSIC.c

    r35768 r36085  
    2020   * note that a Sersic model is usually defined in terms of R_e, the half-light radius.  This
    2121     construction does not include a factor of 2 in the X^2 term, etc, like for a Gaussian.
    22      Conversion from SXX, SYY, SXY to R_major, R_minor, theta can be done by using setting:
     22     Conversion from SXX, SYY, SXY to R_major, R_minor, theta can be done by using:
    2323     shape.sx = SXX / sqrt(2), shape.sy = SYY / sqrt(2), shape.sxy = SXY, then calling
    2424     psEllipseShapeToAxes, and multiplying the values of axes.major, axes.minor by sqrt(2)
    … …  
    5555#include "pmPSFtry.h"
    5656#include "pmDetections.h"
     57#include "pmModel_CentralPixel.h"
    5758
    5859#include "pmModel_SERSIC.h"
    … …  
    7475
    7576// Lax parameter limits
    76 static float paramsMinLax[] = { -1.0e3, 1.0e-2, -100, -100, 0.001, 0.001, -1.0, 0.05 };
    77 static float paramsMaxLax[] = { 1.0e5, 1.0e8, 1.0e4, 1.0e4, 100, 100, 1.0, 4.0 };
     77static float paramsMinLax[] = { -1.0e3, 1.0e-2, -100, -100, 0.001, 0.001, -1.0, 0.1 };
     78static float paramsMaxLax[] = { 1.0e5, 1.0e9, 1.0e5, 1.0e5, 100, 100, 1.0, 1.0 };
    7879
    7980// Moderate parameter limits
    … …  
    8889static float *paramsMinUse = paramsMinLax;
    8990static float *paramsMaxUse = paramsMaxLax;
    90 static float betaUse[] = { 1000, 3e6, 5, 5, 1.0, 1.0, 0.5, 2.0 };
     91static float betaUse[] = { 2, 3e6, 5, 5, 10.0, 10.0, 0.5, 1.0};
    9192
    9293static bool limitsApply = true;         // Apply limits?
    9394
    94 # include "pmModel_SERSIC.CP.h"
     95// # include "pmModel_SERSIC.CP.h"
    9596
    9697psF32 PM_MODEL_FUNC (psVector *deriv,
    … …  
    111112    psAssert (z >= 0, "do not allow negative z values in model");
    112113
    113     float index = 0.5 / PAR[PM_PAR_7];
    114     float par7 = PAR[PM_PAR_7];
    115     float bn = 1.9992*index - 0.3271;
    116     float Io = exp(bn);
    117 
    118     psF32 f2 = bn*pow(z,par7);
    119     psF32 f1 = Io*exp(-f2);
     114    float Sindex = 0.5 / PAR[PM_PAR_7];
     115    float kappa = pmSersicKappa (Sindex);
     116
     117    float q = kappa*pow(z,PAR[PM_PAR_7]);
     118    psF32 f0 = exp(-q);
     119
     120    assert (isfinite(q));
    120121
    121122    psF32 radius = hypot(X, Y);
    122     if (radius < 1.0) {
    123 
    124         // ** use bilinear interpolation to the given location from the 4 surrounding pixels centered on the object center
    125 
    126         // first, use Rmajor and index to find the central pixel flux (fraction of total flux)
    127         psEllipseAxes axes;
    128         pmModelParamsToAxes (&axes, PAR[PM_PAR_SXX], PAR[PM_PAR_SXY], PAR[PM_PAR_SYY], true);
    129 
    130         // get the central pixel flux from the lookup table
    131         float xPix = (axes.major - centralPixelXo) / centralPixeldX;
    132         xPix = PS_MIN (PS_MAX(xPix, 0), centralPixelNX - 1);
    133         float yPix = (index - centralPixelYo) / centralPixeldY;
    134         yPix = PS_MIN (PS_MAX(yPix, 0), centralPixelNY - 1);
    135 
    136         // the integral of a Sersic has an analytical form as follows:
    137         float logGamma = lgamma(2.0*index);
    138         float bnFactor = pow(bn, 2.0*index);
    139         float norm = 2.0 * M_PI * PS_SQR(axes.major) * index * exp(bn) * exp(logGamma) / bnFactor;
    140 
    141         // XXX interpolate to get the value
    142         // XXX for the moment, just integerize
    143         // XXX I need to multiply by the integrated flux to get the flux in the central pixel
    144         float Vcenter = centralPixel[(int)yPix][(int)xPix] * norm;
    145        
    146         float px1 = 1.0 / PAR[PM_PAR_SXX];
    147         float py1 = 1.0 / PAR[PM_PAR_SYY];
    148         float z10 = PS_SQR(px1);
    149         float z01 = PS_SQR(py1);
    150 
    151         // which pixels do we need for this interpolation?
    152         // (I do not keep state information, so I don't know anything about other evaluations of nearby pixels...)
    153         if ((X >= 0) && (Y >= 0)) {
    154             float z11 = z10 + z01 + PAR[PM_PAR_SXY]; // X * Y positive
    155             float V00 = Vcenter;
    156             float V10 = Io*exp(-bn*pow(z10,par7));
    157             float V01 = Io*exp(-bn*pow(z01,par7));
    158             float V11 = Io*exp(-bn*pow(z11,par7));
    159             f1 = interpolatePixels(V00, V10, V01, V11, X, Y);
    160         }
    161         if ((X < 0) && (Y >= 0)) {
    162             float z11 = z10 + z01 - PAR[PM_PAR_SXY]; // X * Y negative
    163             float V00 = Io*exp(-bn*pow(z10,par7));
    164             float V10 = Vcenter;
    165             float V01 = Io*exp(-bn*pow(z11,par7));
    166             float V11 = Io*exp(-bn*pow(z01,par7));
    167             f1 = interpolatePixels(V00, V10, V01, V11, (1.0 + X), Y);
    168         }
    169         if ((X >= 0) && (Y < 0)) {
    170             float z11 = z10 + z01 - PAR[PM_PAR_SXY]; // X * Y negative
    171             float V00 = Io*exp(-bn*pow(z01,par7));
    172             float V10 = Io*exp(-bn*pow(z11,par7));
    173             float V01 = Vcenter;
    174             float V11 = Io*exp(-bn*pow(z10,par7));
    175             f1 = interpolatePixels(V00, V10, V01, V11, X, (1.0 + Y));
    176         }
    177         if ((X < 0) && (Y < 0)) {
    178             float z11 = z10 + z01 + PAR[PM_PAR_SXY]; // X * Y positive
    179             float V00 = Io*exp(-bn*pow(z11,par7));
    180             float V10 = Io*exp(-bn*pow(z10,par7));
    181             float V01 = Io*exp(-bn*pow(z01,par7));
    182             float V11 = Vcenter;
    183             f1 = interpolatePixels(V00, V10, V01, V11, (1.0 + X), (1.0 + Y));
    184         }
    185     }   
    186 
    187     psF32 z0 = PAR[PM_PAR_I0]*f1;
    188     psF32 f0 = PAR[PM_PAR_SKY] + z0;
    189 
    190     if (!isfinite(z0)) {
    191         fprintf(stderr, "z0 is not finite for %f %f %f %f %f.  Parameters: \n", X, Y, radius, z, f1);
     123    if (radius <= 1.5) {
     124        // Nsub ~ 10*index^2 + 1
     125        psEllipseAxes axes = pmPSF_ModelToAxes(PAR, pmModelClassGetType ("PS_MODEL_SERSIC"));
     126        int Nsub = 2 * ((int)(6.0*Sindex / axes.minor)) + 1;
     127        Nsub = PS_MIN (Nsub, 121);
     128        Nsub = PS_MAX (Nsub, 11);
     129        f0 = pmModelCP_SersicSubpix (X, Y, PAR[PM_PAR_SXX], PAR[PM_PAR_SXY], PAR[PM_PAR_SYY], Sindex, Nsub);
     130    }
     131    if (!isfinite(f0)) {
     132        fprintf(stderr, "f0 is not finite for %f %f %f %f %f.  Parameters: \n", X, Y, radius, z, q);
    192133        fprintf(stderr, "%f %f %f %f %f %f %f %f\n", PAR[0], PAR[1], PAR[2], PAR[3], PAR[4],
    193134            PAR[5], PAR[6], PAR[7]);
    194135    }
    195 
    196     assert (isfinite(f2));
     136    assert (isfinite(f0));
     137
     138    psF32 f1 = PAR[PM_PAR_I0]*f0;
     139    psF32 f = PAR[PM_PAR_SKY] + f1;
     140
    197141    assert (isfinite(f1));
    198     assert (isfinite(z0));
    199     assert (isfinite(f0));
     142    assert (isfinite(f));
    200143
    201144    if (deriv != NULL) {
    … …  
    203146
    204147        dPAR[PM_PAR_SKY]  = +1.0;
    205         dPAR[PM_PAR_I0]   = +f1;
    206 
    207         // gradient is infinite for z = 0; saturate at z = 0.01
    208         psF32 z1 = (z < 0.01) ? z0*bn*par7*pow(0.01,par7 - 1.0) : z0*bn*par7*pow(z,par7 - 1.0);
    209 
    210         dPAR[PM_PAR_7]    = (z < 0.01) ? -z0*pow(0.01,par7)*log(0.01) : -z0*f2*log(z);
    211         dPAR[PM_PAR_7]   *= 3.0;
    212 
    213         assert (isfinite(z1));
     148        dPAR[PM_PAR_I0]   = +f0;
     149
     150        if (z > 0.01) {
     151          float z1 = f1*kappa*PAR[PM_PAR_7]*pow(z,PAR[PM_PAR_7]-1.0);
     152          dPAR[PM_PAR_XPOS] = +1.0*z1*(2.0*px + Y*PAR[PM_PAR_SXY]);
     153          dPAR[PM_PAR_YPOS] = +1.0*z1*(2.0*py + X*PAR[PM_PAR_SXY]);
     154          dPAR[PM_PAR_SXX]  = +2.0*z1*px*px/PAR[PM_PAR_SXX];
     155          dPAR[PM_PAR_SYY]  = +2.0*z1*py*py/PAR[PM_PAR_SYY];
     156          dPAR[PM_PAR_SXY]  = -1.0*z1*X*Y;
     157          dPAR[PM_PAR_7]    = -1.0*f1*q*log(z);
     158        } else {
     159          // gradient -> 0 for z -> 0, but has undef form
     160          float z1 = f1*kappa*PAR[PM_PAR_7]*pow(z,PAR[PM_PAR_7]);
     161          dPAR[PM_PAR_XPOS] = +1.0*z1*(2.0/PAR[PM_PAR_SXX] + PAR[PM_PAR_SXY]);
     162          dPAR[PM_PAR_YPOS] = +1.0*z1*(2.0/PAR[PM_PAR_SYY] + PAR[PM_PAR_SXY]);
     163          dPAR[PM_PAR_SXX]  = +2.0*z1*px/PAR[PM_PAR_SXX]/PAR[PM_PAR_SXX];
     164          dPAR[PM_PAR_SYY]  = +2.0*z1*py/PAR[PM_PAR_SYY]/PAR[PM_PAR_SYY];
     165          dPAR[PM_PAR_SXY]  = -1.0*z1;
     166          // dPAR[PM_PAR_7]    = -1.0*f1*q*log(z + 0.0001);
     167          dPAR[PM_PAR_7]    = -1.0*f1*q*log(z + 0.0001); // factor of 16 to reduce the gain
     168        }
    214169        assert (isfinite(dPAR[PM_PAR_7]));
    215 
    216         dPAR[PM_PAR_XPOS] = +1.0*z1*(2.0*px/PAR[PM_PAR_SXX] + Y*PAR[PM_PAR_SXY]);
    217         dPAR[PM_PAR_YPOS] = +1.0*z1*(2.0*py/PAR[PM_PAR_SYY] + X*PAR[PM_PAR_SXY]);
    218         dPAR[PM_PAR_SXX]  = +2.0*z1*px*px/PAR[PM_PAR_SXX]; // XXX : increase drag?
    219         dPAR[PM_PAR_SYY]  = +2.0*z1*py*py/PAR[PM_PAR_SYY];
    220         dPAR[PM_PAR_SXY]  = -1.0*z1*X*Y;
    221     }
    222     return (f0);
     170    }
     171    return (f);
    223172}
    224173
    … …  
    370319    psEllipseAxes axes;
    371320    pmModelParamsToAxes (&axes, PAR[PM_PAR_SXX], PAR[PM_PAR_SXY], PAR[PM_PAR_SYY], true);
    372     float AspectRatio = axes.minor / axes.major;
    373 
    374     float index = 0.5 / PAR[PM_PAR_7];
    375     float bn = 1.9992*index - 0.3271;
    376 
    377     // the integral of a Sersic has an analytical form as follows:
    378     float logGamma = lgamma(2.0*index);
    379     float bnFactor = pow(bn, 2.0*index);
    380     float norm = 2.0 * M_PI * PS_SQR(axes.major) * index * exp(bn) * exp(logGamma) / bnFactor;
    381    
    382     psF64 Flux = PAR[PM_PAR_I0] * norm * AspectRatio;
    383 
    384     return(Flux);
     321
     322    float Sindex = 0.5 / PAR[PM_PAR_7];
     323    float norm = pmSersicNorm (Sindex);
     324
     325    float flux = PAR[PM_PAR_I0] * 2.0 * M_PI * axes.major * axes.minor * norm;
     326
     327    return(flux);
    385328}
    386329
    … …  
    401344    pmModelParamsToAxes (&axes, PAR[PM_PAR_SXX], PAR[PM_PAR_SXY], PAR[PM_PAR_SYY], true);
    402345
     346    float Sindex = 0.5 / PAR[PM_PAR_7];
     347    float kappa = pmSersicKappa (Sindex);
     348
    403349    // f = Io exp(-z^n) -> z^n = ln(Io/f)
    404     psF64 zn = log(PAR[PM_PAR_I0] / flux);
    405     psF64 radius = axes.major * sqrt (2.0) * pow(zn, 0.5 / PAR[PM_PAR_7]);
     350    psF64 zn = log(PAR[PM_PAR_I0] / flux) / kappa;
     351    psF64 radius = axes.major * pow(zn, Sindex);
    406352
    407353    psAssert (isfinite(radius), "fix this code: radius should not be nan for Io = %f, flux = %f, major = %f (%f, %f, %f), par 7 = %f",
Note: See TracChangeset for help on using the changeset viewer.