Skip to content

I
imagej-elphel
Commit 3476fe17 authored by Andrey Filippov's avatar Andrey Filippov
Browse files
Options

Generating mask and result rms error for different cutoff disparities

parent b80e5500
Hide whitespace changes
Inline Side-by-side
Showing with 125 additions and 12 deletions
src/main/java/MLStats.java
View file @ 3476fe17
......@@ -64,9 +64,15 @@ public class MLStats {
double strength_max_drop = 1.0; //
double strength_max_clip = 0.9; //
boolean normalize = true;
double master_weight_power = 1.0;
double master_weight_floor = 0.08;
double disparity_outlier = 1.0;
double master_weight_power = 1.0;
double master_weight_floor = 0.08;
double disparity_outlier = 1.0;
double pre_log_offs = 0.01; // add before log to avoid -infinity
double log_sigma = 2.00; // blur logarithm of the histogram (in bins)
double mask_threshold = 0.25; // relative tile population
double log_sigma_d = 2.00; // blur logarithm of the histogram in disparity pixels
double log_sigma_s = 0.01; // blur logarithm of the histogram in strength units
int result_disparity_step = 10; // bins
String mask = ".*-DSI_COMBO\\.tiff";
......@@ -88,6 +94,14 @@ public class MLStats {
gd.addNumericField("Master weight floor", master_weight_floor, 3);
gd.addNumericField("Ignore tiles with disparity difference higher", disparity_outlier, 3);
gd.addNumericField("Mask: add before log to avoid -infinity", pre_log_offs, 3);
gd.addNumericField("Mask: blur logarithm of the histogram", log_sigma, 3);
gd.addNumericField("Mask: threshold (relative tile population)", mask_threshold, 3);
gd.addNumericField("Blur logarithm of the histogram in disparity pixels", log_sigma_d, 3);
gd.addNumericField("Blur logarithm of the histogram in strength units", log_sigma_s, 3);
gd.addNumericField("Report RMS for each disaprity bins", result_disparity_step, 0);
gd.showDialog ();
if (gd.wasCanceled()) return false;
mask = gd.getNextString();
......@@ -105,6 +119,15 @@ public class MLStats {
master_weight_power = gd.getNextNumber();
master_weight_floor = gd.getNextNumber();
disparity_outlier = gd.getNextNumber();
pre_log_offs = gd.getNextNumber();
log_sigma = gd.getNextNumber();
mask_threshold = gd.getNextNumber();
log_sigma_d = gd.getNextNumber();
log_sigma_s = gd.getNextNumber();
result_disparity_step = (int) gd.getNextNumber();
// get list of all files:
System.out.println("File mask = "+mask);
......@@ -135,6 +158,7 @@ public class MLStats {
double disparity_offs = disparity_min_clip - disparity_step/2; // last and first bin that include clip will be 0.5 width
double strength_offs = strength_min_clip - strength_step/2; // last and first bin that include clip will be 0.5 width
int total_tiles_used = 0;
int nfile = 0;
for (Path p:files) {
ImagePlus imp_dsi=new ImagePlus(p.normalize().toString());
ImageStack dsi_stack= imp_dsi.getStack();
......@@ -219,11 +243,14 @@ public class MLStats {
}
}
}
System.out.println(p.getFileName()+": "+nut+" useful tiles counted");
System.out.println((++nfile)+": "+p.getFileName()+": "+nut+" useful tiles counted");
total_tiles_used += nut;
}
System.out.println("Total number of useful tiles: "+total_tiles_used);
double [][] hist_double = new double [3][disparity_bins*strength_bins];
System.out.println("Total number of useful tiles: "+total_tiles_used+ " of "+nfile+" files");
String [] titles = {"histogram", "histogram_ideal", "disp_err","disp_err9", "masked_err","masked_err9"};
double [][] hist_double = new double [titles.length][disparity_bins*strength_bins];
double scale = 1.0;
if (normalize) {
scale *= (1.0* disparity_bins * strength_bins) / total_tiles_used;
......@@ -233,18 +260,104 @@ public class MLStats {
int dbin = nTile % disparity_bins;
int sbin = nTile / disparity_bins;
hist_double[0][nTile] = scale * hist[dbin][sbin];
if (ds_error[dbin][sbin][1] > 0.0) {
hist_double[1][nTile] = Math.sqrt(ds_error[dbin][sbin][0]/ds_error[dbin][sbin][1]);
}
// create a mask of relatively frequent d/s cells
// double pre_log_offs = 0.01; // add before log to avoid -infinity
// double log_sigma = 2.00; // blur logarithm of the histogram
// double mask_threshold = 0.25; // relative tile population
double [] mask_calc = new double [disparity_bins*strength_bins]; // hist_double[0].clone();
for (int nTile = 0; nTile < mask_calc.length; nTile++ ) {
mask_calc[nTile] = Math.log( hist_double[0][nTile] + pre_log_offs);
}
if (log_sigma > 0.0) {
(new DoubleGaussianBlur()).blurDouble(
mask_calc,
disparity_bins, // int width,
strength_bins, // int height,
log_sigma, // double sigmaX,
log_sigma, // double sigmaY,
0.01); // double accuracy)
}
double log_threshold = Math.log(mask_threshold+pre_log_offs);
boolean [] ds_mask = new boolean [disparity_bins*strength_bins];
for (int nTile = 0; nTile < mask_calc.length; nTile++ ) {
ds_mask[nTile] = mask_calc[nTile] >= log_threshold;
}
// log_sigma_d = gd.getNextNumber();
// log_sigma_s = gd.getNextNumber();
// double disparity_step = (disparity_max_clip - disparity_min_clip) / disparity_bins;
// double strength_step = (strength_max_clip - strength_min_clip) / strength_bins;
(new DoubleGaussianBlur()).blurDouble(
mask_calc,
disparity_bins, // int width,
strength_bins, // int height,
log_sigma_d/disparity_step, // double sigmaX,
log_sigma_s/disparity_step, // double sigmaY,
0.01); // double accuracy)
for (int nTile = 0; nTile < mask_calc.length; nTile++ ) {
if (!ds_mask[nTile]) {
mask_calc[nTile] = 0.0;
} else {
hist_double[1][nTile] = Double.NaN;
mask_calc[nTile] = Math.exp(mask_calc[nTile]) - log_threshold;
}
if (ds_error[dbin][sbin][3] > 0.0) {
hist_double[2][nTile] = Math.sqrt(ds_error[dbin][sbin][2]/ds_error[dbin][sbin][3]);
}
for (int nTile = 0; nTile < hist_double[0].length; nTile++) {
int dbin = nTile % disparity_bins;
int sbin = nTile / disparity_bins;
hist_double[1][nTile] = mask_calc[nTile];
if (ds_error[dbin][sbin][2] > 0.0) {
hist_double[2][nTile] = Math.sqrt(ds_error[dbin][sbin][0]/ds_error[dbin][sbin][1]);
} else {
hist_double[2][nTile] = Double.NaN;
}
if (ds_error[dbin][sbin][3] > 0.0) {
hist_double[3][nTile] = Math.sqrt(ds_error[dbin][sbin][2]/ds_error[dbin][sbin][3]);
} else {
hist_double[3][nTile] = Double.NaN;
}
if (ds_mask[nTile] && (ds_error[dbin][sbin][2] > 0.0)) {
hist_double[4][nTile] = Math.sqrt(ds_error[dbin][sbin][0]/ds_error[dbin][sbin][1]);
} else {
hist_double[4][nTile] = Double.NaN;
}
if (ds_mask[nTile] && (ds_error[dbin][sbin][3] > 0.0)) {
hist_double[5][nTile] = Math.sqrt(ds_error[dbin][sbin][2]/ds_error[dbin][sbin][3]);
} else {
hist_double[5][nTile] = Double.NaN;
}
}
String [] titles = {"histogram", "disp_err","disp_err9"};
// result_disparity_step
// Calculate weighted by both master and strength bin (rig strength), report for each result_disparity_step disparity bins (running totals)
double sew=0.0, sw=0.0, sew9 = 0.0, sw9 = 0.0;
for (int disp0 = 0; disp0 < disparity_bins; disp0 += result_disparity_step) {
int dbin = disp0;
for (; (dbin < (disp0 + result_disparity_step)) && (dbin < disparity_bins); dbin ++) {
for (int sbin = 0; sbin < strength_bins; sbin++) {
int nTile = dbin+sbin*disparity_bins;
if (ds_mask[nTile]) {
sew += ds_error[dbin][sbin][0];
sw += ds_error[dbin][sbin][1];
sew9 += ds_error[dbin][sbin][2];
sw9 += ds_error[dbin][sbin][3];
}
}
}
double run_disp = disparity_offs + disparity_step * dbin;
double rms = Math.sqrt(sew/sw);
double rms9 = Math.sqrt(sew9/sw9);
System.out.println(String.format("disparity: %7.3f pix rms= %5.3f rms9= %5.3f", run_disp, rms, rms9));
}
ImagePlus imp = (new showDoubleFloatArrays()).makeArrays(
hist_double,
disparity_bins,
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment