Skip to content

I
imagej-elphel
Commit 0ab422e2 authored by Andrey Filippov's avatar Andrey Filippov
Browse files
Options

removing foreground bridges

parent 971dd284
Hide whitespace changes
Inline Side-by-side
Showing with 889 additions and 946 deletions
src/main/java/Conflicts.java
View file @ 0ab422e2
......@@ -284,7 +284,7 @@ public class Conflicts {
final int stilesY = (tilesY + superTileSize -1)/superTileSize;
final int nStiles = stilesX * stilesY;
final int [][][] conflicts = new int [st.getPlanes().length][][];
final TileSurface.TileNeibs tnSurface = st.getTileSurface().new TileNeibs(stilesX, stilesY);
final TileNeibs tnSurface = new TileNeibs(stilesX, stilesY);
final Thread[] threads = ImageDtt.newThreadArray(st.getTileProcessor().threadsMax);
final AtomicInteger ai = new AtomicInteger(0);
for (int ithread = 0; ithread < threads.length; ithread++) {
......@@ -343,7 +343,7 @@ public class Conflicts {
HashMap<Integer,Integer> replacement_tiles, // null is OK
int [][][] replacement_neibs, // null OK if replacement_tiles == null
double [][][] replacement_val_weights,
TileSurface.TileNeibs tnSurface)
TileNeibs tnSurface)
{
TilePlanes.PlaneData [][] planes = st.getPlanes();
// generate conflicts if not provided
......@@ -405,7 +405,7 @@ public class Conflicts {
HashMap<Integer,Integer> replacement_tiles, // null is OK
int [][][] replacement_neibs, // null OK if replacement_tiles == null
ConnectionCosts connectionCosts,
TileSurface.TileNeibs tnSurface)
TileNeibs tnSurface)
{
TilePlanes.PlaneData [][] planes = st.getPlanes();
// generate conflicts if not provided
......@@ -465,7 +465,7 @@ public class Conflicts {
int nsTile0,
HashMap<Integer,Integer> replacement_tiles, // null is OK - will use only planes data
int [][][] replacement_neibs, // null OK if replacement_tiles == null
TileSurface.TileNeibs tnSurface)
TileNeibs tnSurface)
{
TilePlanes.PlaneData [][] planes = st.getPlanes();
ArrayList<Conflict> conflicts_list= new ArrayList<Conflict>();
......
src/main/java/ConnectionCosts.java
View file @ 0ab422e2
......@@ -28,7 +28,7 @@ import java.util.HashSet;
public class ConnectionCosts {
TilePlanes.PlaneData [][] planes = null;
boolean preferDisparity = false;
TileSurface.TileNeibs tnSurface;
TileNeibs tnSurface;
double orthoWeight;
double diagonalWeight;
double starPwr; // Divide cost by number of connections to this power
......@@ -57,7 +57,7 @@ public class ConnectionCosts {
double starValPwr, // Raise value of each tile before averaging
int steps,
TilePlanes.PlaneData [][] planes,
TileSurface.TileNeibs tnSurface,
TileNeibs tnSurface,
boolean preferDisparity)
{
this.planes = planes;
......@@ -481,7 +481,7 @@ public class ConnectionCosts {
double diagonalWeight,
double starPwr, // Divide cost by number of connections to this power
double starWeightPwr, // Use this power of tile weight when calculating connection cost
TileSurface.TileNeibs tnSurface,
TileNeibs tnSurface,
boolean preferDisparity,
int debugLevel)
{
......@@ -506,7 +506,7 @@ public class ConnectionCosts {
double diagonalWeight,
double starPwr, // Divide cost by number of connections to this power
double starWeightPwr, // Use this power of tile weight when calculating connection cost
TileSurface.TileNeibs tnSurface,
TileNeibs tnSurface,
boolean preferDisparity,
int debugLevel)
{
......@@ -566,7 +566,7 @@ public class ConnectionCosts {
double diagonalWeight,
double starPwr, // Divide cost by number of connections to this power
double starWeightPwr, // Use this power of tile weight when calculating connection cost
TileSurface.TileNeibs tnSurface,
TileNeibs tnSurface,
boolean preferDisparity,
int debugLevel)
{
......@@ -594,7 +594,7 @@ public class ConnectionCosts {
double diagonalWeight,
double starPwr, // Divide cost by number of connections to this power
double starWeightPwr, // Use this power of tile weight when calculating connection cost
TileSurface.TileNeibs tnSurface,
TileNeibs tnSurface,
boolean preferDisparity,
int debugLevel)
{
......
src/main/java/LinkPlanes.java
View file @ 0ab422e2
......@@ -813,6 +813,17 @@ public class LinkPlanes {
ImageDtt.startAndJoin(threads);
}
/**
* Merge the supertile planes with agreeing neighbors, non-exclusively (no considering other planes
* of the same supertile. Start with the best fit, then goes to lower quality, until the individual
* merge quality falls below scaled quality of the best, pre-set minimum or the merged plane becomes
* too thick
* Separately calculates merged weighted plane and with equal weights of the neighbors
* @param planes array of plane instances for the same supertile
* @param debugLevel
* @param dbg_X
* @param dbg_Y
*/
public void setNonExclusive(
final TilePlanes.PlaneData [][] planes,
final int debugLevel,
......@@ -834,7 +845,7 @@ public class LinkPlanes {
// final int debug_stile = 9 * stilesX + 26;
final int debug_stile = dbg_Y * stilesX + dbg_X;
final Thread[] threads = ImageDtt.newThreadArray(st.tileProcessor.threadsMax);
final Thread[] threads = ImageDtt.newThreadArray((debugLevel > 1)? 1 : st.tileProcessor.threadsMax);
final AtomicInteger ai = new AtomicInteger(0);
for (int ithread = 0; ithread < threads.length; ithread++) {
......@@ -2426,7 +2437,7 @@ public class LinkPlanes {
final int stilesX = (tilesX + superTileSize -1)/superTileSize;
final int stilesY = (tilesY + superTileSize -1)/superTileSize;
final int nStiles = stilesX * stilesY;
final TileSurface.TileNeibs tnSurface = st.tileSurface.new TileNeibs(stilesX, stilesY);
final TileNeibs tnSurface = new TileNeibs(stilesX, stilesY);
final Thread[] threads = ImageDtt.newThreadArray(st.tileProcessor.threadsMax);
final AtomicInteger ai = new AtomicInteger(0);
for (int ithread = 0; ithread < threads.length; ithread++) {
......@@ -2441,7 +2452,7 @@ public class LinkPlanes {
starValPwr, //double starValPwr, // Raise value of each tile before averaging
steps, // int steps,
planes, // TilePlanes.PlaneData [][] planes,
tnSurface, // TileSurface.TileNeibs tnSurface,
tnSurface, // TileNeibs tnSurface,
preferDisparity); // boolean preferDisparity)
int [] mod_supertiles = new int[1];
for (int nsTile = ai.getAndIncrement(); nsTile < nStiles; nsTile = ai.getAndIncrement()) {
......@@ -2484,7 +2495,7 @@ public class LinkPlanes {
// final int tileSize = tileProcessor.getTileSize();
final int stilesX = (tilesX + superTileSize -1)/superTileSize;
final int stilesY = (tilesY + superTileSize -1)/superTileSize;
final TileSurface.TileNeibs tnSurface = st.tileSurface.new TileNeibs(stilesX, stilesY);
final TileNeibs tnSurface = new TileNeibs(stilesX, stilesY);
final Thread[] threads = ImageDtt.newThreadArray(st.tileProcessor.threadsMax);
final AtomicInteger ai = new AtomicInteger(0);
for (int ithread = 0; ithread < threads.length; ithread++) {
......@@ -2499,7 +2510,7 @@ public class LinkPlanes {
starValPwr, //double starValPwr, // Raise value of each tile before averaging
steps, // int steps,
planes, // TilePlanes.PlaneData [][] planes,
tnSurface, // TileSurface.TileNeibs tnSurface,
tnSurface, // TileNeibs tnSurface,
preferDisparity); // boolean preferDisparity)
int [] supertiles = new int[1];
for (int isTile = ai.getAndIncrement(); isTile < mod_supertiles.length; isTile = ai.getAndIncrement()) {
......
src/main/java/SuperTiles.java
View file @ 0ab422e2
......@@ -1025,147 +1025,6 @@ public class SuperTiles{
return bgTileDispStrength;
}
public void processPlanes(
final boolean [] selected, // or null
final double min_disp,
final boolean invert_disp, // use 1/disparity
final double plDispNorm,
final int debugLevel)
{
final int tilesX = tileProcessor.getTilesX();
final int tilesY = tileProcessor.getTilesY();
final int superTileSize = tileProcessor.getSuperTileSize();
final int tileSize = tileProcessor.getTileSize();
final double [] disparity = cltPass3d.getDisparity();
final double [] strength = cltPass3d.getStrength();
final int stilesX = (tilesX + superTileSize -1)/superTileSize;
final int stilesY = (tilesY + superTileSize -1)/superTileSize;
final int nStiles = stilesX * stilesY;
final Thread[] threads = ImageDtt.newThreadArray(tileProcessor.threadsMax);
final AtomicInteger ai = new AtomicInteger(0);
final int st_start = -superTileSize/2;
final int superTileSize2 = 2 * superTileSize;
final double [][] lapWeight = getLapWeights();
final int len2 = superTileSize2*superTileSize2;
final double [] double_zero = new double [len2];
final boolean [] boolean_zero = new boolean [len2];
final int debug_stile = 18 * stilesX + 25;
for (int ithread = 0; ithread < threads.length; ithread++) {
threads[ithread] = new Thread() {
public void run() {
TilePlanes tpl = new TilePlanes(tileSize,superTileSize);
double [] stDisparity = new double [superTileSize2*superTileSize2];
double [] stStrength = new double [superTileSize2*superTileSize2];
boolean [] stSel = new boolean [superTileSize2*superTileSize2];
for (int nsTile = ai.getAndIncrement(); nsTile < nStiles; nsTile = ai.getAndIncrement()) {
int stileY = nsTile / stilesX;
int stileX = nsTile % stilesX;
double sw = 0.0; // sum weights
double [] hist = new double [numBins];
int tY0 = stileY * superTileSize + st_start;
int tX0 = stileX * superTileSize + st_start;
System.arraycopy(double_zero, 0, stDisparity, 0, len2);
System.arraycopy(double_zero, 0, stStrength, 0, len2);
System.arraycopy(boolean_zero, 0, stSel, 0, len2);
for (int tY = 0; tY < superTileSize2; tY++){
int tileY = tY0 +tY;
if ((tileY >= 0) && (tileY < tilesY)) {
for (int tX = 0; tX < superTileSize2; tX++){
int tileX = tX0 +tX;
if ((tileX >= 0) && (tileX < tilesX)) {
int indx = tileY*tilesX + tileX;
double d = disparity[indx];
if (!Double.isNaN(d) && (d >= min_disp) &&((selected == null) || selected[indx])){
if (invert_disp){
d = 1.0/d;
}
double w = strength[indx] - strength_floor;
if (w > 0.0){
if (strength_pow != 1.0) w = Math.pow(w, strength_pow);
w *= lapWeight[tY][tX];
int indx_out = tY * superTileSize2 + tX;
stDisparity[indx_out] = d;
stStrength[indx_out] = w;
stSel[indx_out] = true;
sw +=w;
}
}
}
}
}
}
if (sw >0){
// int dl = ((nsTile >= debug_stile-1) && (nsTile <= debug_stile+1) ) ? 1 : 0;
// int dl = (stileY == 17) ? 1 : 0;
int dl = (stileY >= 0) ? 1 : 0;
double [][][] rslt = tpl.getCovar(
stDisparity,
stStrength,
stSel,
plDispNorm,
0); // dl); // debugLevel);
if (dl > 0) {
int numPoints = (int) rslt[2][0][2];
double kz = rslt[2][0][1];
double swc = rslt[2][0][0];
double [] szxy = rslt[2][1];
double [][] eig_val = rslt[0];
double [][] eig_vect = rslt[1];
if (swc > 1.0) {
System.out.println("Processing planes, nsTile="+nsTile+", stileX="+stileX+", stileY="+stileY+", numPoints="+numPoints+
", kz = "+kz+", sw = "+sw+ ", swc = "+swc+ ", center=["+szxy[0]+","+szxy[1]+","+szxy[2]+"]"+
", eig_val = {"+eig_val[0][0]+","+eig_val[1][1]+","+eig_val[2][2]+"}"+
", eig_vect[0] = {"+eig_vect[0][0]+","+eig_vect[1][0]+","+eig_vect[2][0]+"}");
}
}
/*
double [][][] rslt = {
eig.getD().getArray(),
eig.getV().getArray(),
{
{sw,kz},
{swz, swx, swy}}};
return rslt;
*/
if (dl > 1) {
System.out.println("Processing planes with average disparity");
double ss = 0.0, sd = 0.0;
for (int i = 0; i < stDisparity.length; i++){
if (!Double.isNaN(stDisparity[i]) && stSel[i]){
ss += stStrength[i];
sd += stStrength[i] * stDisparity[i];
}
}
if (ss > 0) {
sd /= ss;
}
for (int i = 0; i < stDisparity.length; i++){
stDisparity[i] = sd;
}
tpl.getCovar(
stDisparity,
stStrength,
stSel,
plDispNorm,
dl); // debugLevel);
}
}
}
}
};
}
ImageDtt.startAndJoin(threads);
}
public double corrMaxEigen(
double maxEigen,
......@@ -1766,13 +1625,38 @@ public class SuperTiles{
return plane_selections;
}
// use both horizontal and const disparity tiles to create tile clusters
// Add max_diff (maximal disparity difference while extracting initial tile selection) and max_tries (2..3) parameters
// Add separate method to create + remove outliers from all planes (2 different ones)?
// TODO later re-assign pixels according to existing plane parameters
// Sort plane data by center (plane or supertile) disparity
/**
* Use both horizontal and const disparity tiles to create tile clusters
*Add max_diff (maximal disparity difference while extracting initial tile selection) and max_tries (2..3) parameters
* Add separate method to create + remove outliers from all planes (2 different ones)?
* TODO later re-assign pixels according to existing plane parameters
* Sort plane data by center (plane or supertile) disparity
*
* @param growSelection
* @param stMeasSel
* @param plDispNorm
* @param plMinPoints
* @param plPreferDisparity
* @param geometryCorrection
* @param correct_distortions
* @param smplMode
* @param smplSide
* @param smplNum
* @param smplRms
* @param bin_blur_hor
* @param bin_blur_vert
* @param max_diff_hor
* @param max_diff_vert
* @param max_tries
* @param smallDiff
* @param highMix
* @param world_hor
* @param debugLevel
* @param dbg_X
* @param dbg_Y
* @return
*/
public boolean [][][][] initialDiscriminateTiles(
final int growSelection, // grow initial selection before processing
final int stMeasSel, // = 1; // Select measurements for supertiles : +1 - combo, +2 - quad +4 - hor +8 - vert
......@@ -2057,16 +1941,37 @@ public class SuperTiles{
}
return planes_selections;
}
public TilePlanes.PlaneData [][] createPlanesFromSelectionsOld(
final boolean [][][][] plane_selections, // = new boolean [nStiles][][][]; // num_tiles
final double [][][][] disp_strength,
/**
* Re-assign tiles to the planes according to fitted planes. Scans each known plane parallel with specified
* size steps and specified number of steps, maximizing overall fit quality.
* Total weight of the fitted tiles? RMS?
* @param planes per-supertile, per plane array of plane data instances. Should have
* nonexclusiveStar and nonexclusiveStarEq calculated
* @param plDispNorm - normalization of the measured disparity precision - closer objects will have disparity
* difference proportionally reduced
* @param plMinPoints minimal number of points in a plane
* @param plPreferDisparity kept for historical reasons - when true, select disparity-most vector even if it has higher eigenvalue
* @param geometryCorrection GeometryCorrection instance to use
* @param correct_distortions correct geometrical distortions when converting to/from world coordinates
* @param smplMode sample mode
* @param smplSide sample side for averaging/filtering tile disparities
* @param smplNum number of best samples used fro averaging
* @param smplRms maximal sample disparity rms to consider sample valid
* @param max_disp_diff maximal disparity difference from the planes
* @param disp_step disparity scan step when searching for optimal fit
* @param steps_per_plane number of disparity steps to try around initial plane
* @param mode which of the calculated planes to use - 0 - weighted, 1 - equalized, 2 heaviest of the two, weighted combination
* @param debugLevel debug level
* @param dbg_X supertile X for elevated debug level
* @param dbg_Y supertile X for elevated debug level
* @return per-tile, per plane, per measurement layer (type of correlation - combo, 4, hor, vert), per tile selections of
* filtered tiles
*/
public boolean [][][][] refineDiscriminateTiles(
final TilePlanes.PlaneData [][] planes,
final double plDispNorm,
final int plMinPoints, // = 5; // Minimal number of points for plane detection
final double plTargetEigen, // = 0.1; // Remove outliers until main axis eigenvalue (possibly scaled by plDispNorm) gets below
final double plFractOutliers, // = 0.3; // Maximal fraction of outliers to remove
final int plMaxOutliers, // = 20; // Maximal number of outliers to remove
// final double plVertWors, // = 1.5 // if rotating plane vertical does not increase 'eigenvalue' more, use vertical
final boolean plPreferDisparity, // Always start with disparity-most axis (false - lowest eigenvalue)
final GeometryCorrection geometryCorrection,
final boolean correct_distortions,
......@@ -2075,185 +1980,23 @@ public class SuperTiles{
final int smplSide, // = 2; // Sample size (side of a square)
final int smplNum, // = 3; // Number after removing worst
final double smplRms, // = 0.1; // Maximal RMS of the remaining tiles in a sample
final double max_disp_diff, // maximal disparity difference from the plane to consider tile
final double disp_step, // parallel move known planes around original know value for the best overall fit
final int steps_per_plane, // number of steps for each plane to search for the best fit
final int mode, // 0 - weighted, 1 - equalized, 2 - best, 3 - combined
final int debugLevel,
final int dbg_X,
final int dbg_Y)
{
final int tilesX = tileProcessor.getTilesX();
final int tilesY = tileProcessor.getTilesY();
final int superTileSize = tileProcessor.getSuperTileSize();
final int tileSize = tileProcessor.getTileSize();
final int stilesX = (tilesX + superTileSize -1)/superTileSize;
final int stilesY = (tilesY + superTileSize -1)/superTileSize;
final int nStiles = stilesX * stilesY;
final Thread[] threads = ImageDtt.newThreadArray(tileProcessor.threadsMax);
final AtomicInteger ai = new AtomicInteger(0);
final TilePlanes.PlaneData [][] result_planes = new TilePlanes.PlaneData[nStiles][];
// this.planes = new TilePlanes.PlaneData[nStiles][];
final int debug_stile = (debugLevel > -1)? (dbg_Y * stilesX + dbg_X):-1;
// TODO: Remove when promoting PlaneData
final TilePlanes tpl = new TilePlanes(tileSize,superTileSize, geometryCorrection);
for (int ithread = 0; ithread < threads.length; ithread++) {
threads[ithread] = new Thread() {
public void run() {
for (int nsTile = ai.getAndIncrement(); nsTile < nStiles; nsTile = ai.getAndIncrement()) {
if (nsTile == debug_stile){
System.out.println("createPlanesFromSelections(): nsTile="+nsTile);
}
if (plane_selections[nsTile] != null) {
int stileY = nsTile / stilesX;
int stileX = nsTile % stilesX;
int [] sTiles = {stileX, stileY};
int dl = (nsTile == debug_stile) ? 3 : 0;
result_planes[nsTile] = null;
// first make a plane from all tiles
ArrayList<TilePlanes.PlaneData> st_planes = new ArrayList<TilePlanes.PlaneData>();
TilePlanes.PlaneData pd0 = tpl.new PlaneData (
sTiles, // int [] sTileXY,
tileSize, // int tileSize,
geometryCorrection, // GeometryCorrection geometryCorrection,
correct_distortions,
measuredLayers, // MeasuredLayers measuredLayers,
plPreferDisparity); // boolean preferDisparity)
// iterate through all plane selections
for (int ps = 0; ps < plane_selections[nsTile].length; ps++) {
TilePlanes.PlaneData pd = pd0.clone();
boolean OK = (pd.getPlaneFromMeas(
plane_selections[nsTile][ps], // tile_sel, // boolean [][] tile_sel, // null - do not use, {} use all (will be modified)
disp_strength[nsTile],
Double.NaN, // double disp_far, // minimal disparity to select (or NaN)
Double.NaN, // double disp_near, // maximal disparity to select (or NaN)
plDispNorm, // 0.0, // plDispNorm, // double dispNorm, // Normalize disparities to the average if above
0.0, // double min_weight,
plMinPoints, // int min_tiles,
strength_floor, //
strength_pow, // double strength_pow,
// update !
smplMode,
smplSide,
smplNum,
smplRms,
dl) != null); // int debugLevel)
if (OK) {
if (dl > 0) {
if (pd.getWeight() > 1.0) {
System.out.println("Processing subplane["+nsTile+"]["+ps+"]"+
", stileX="+stileX+
", stileY="+stileY+
", numPoints="+ pd.getNumPoints()+
", swc = "+pd.getWeight()+
", center=["+pd.getZxy()[0]+","+pd.getZxy()[1]+","+pd.getZxy()[2]+"]"+
", eig_val = {"+pd.getValues()[0]+","+pd.getValues()[1]+","+pd.getValues()[2]+"}"+
", eig_vect[0] = {"+pd.getVector()[0]+","+pd.getVector()[1]+","+pd.getVector()[2]+"}");
}
}
// now try to remove outliers
int max_outliers = (int) Math.round(pd.getNumPoints() * plFractOutliers);
if (max_outliers > plMaxOutliers) max_outliers = plMaxOutliers;
double targetV = plTargetEigen;
double z0 = pd.getZxy()[0];
if ((plDispNorm > 0.0) && (z0 > plDispNorm)) {
double dd = (plDispNorm + z0)/ plDispNorm; // > 1
targetV *= dd * dd; // > original
}
if (pd.getValues()[0] > targetV) {
OK = pd.removeOutliers( // getPlaneFromMeas should already have run
disp_strength[nsTile],
targetV, // double targetEigen, // target eigenvalue for primary axis (is disparity-dependent, so is non-constant)
max_outliers, // int maxRemoved, // maximal number of tiles to remove (not a constant)
dl); // int debugLevel)
if (!OK) {
continue;
}
if (dl > 0) {
if (pd.getWeight() > 1.0) {
System.out.println("Removed outliers["+nsTile+"]["+ps+"]"+
", stileX="+stileX+
", stileY="+stileY+
", numPoints="+ pd.getNumPoints()+
", swc = "+pd.getWeight()+
", center=["+pd.getZxy()[0]+","+pd.getZxy()[1]+","+pd.getZxy()[2]+"]"+
", eig_val = {"+pd.getValues()[0]+","+pd.getValues()[1]+","+pd.getValues()[2]+"}"+
", eig_vect[0] = {"+pd.getVector()[0]+","+pd.getVector()[1]+","+pd.getVector()[2]+"}");
}
}
}
double [] norm_xyz = pd.getWorldXYZ(
correct_distortions);
st_planes.add(pd);
if (dl > 0) {
System.out.println("World normal["+nsTile+"]["+ps+"] = {"+
norm_xyz[0]+", "+norm_xyz[1]+", "+norm_xyz[2]+"}");
}
// calculate the world planes too
// if (debugLevel > -1){
pd.getWorldPlaneFromMeas(
plane_selections[nsTile][ps], // tile_sel, // boolean [][] tile_sel, // null - do not use, {} use all (will be modified)
disp_strength[nsTile],
Double.NaN, // double disp_far, // minimal disparity to select (or NaN)
Double.NaN, // double disp_near, // maximal disparity to select (or NaN)
plDispNorm, // 0.0, // plDispNorm, // double dispNorm, // Normalize disparities to the average if above
0.0, // double min_weight,
plMinPoints, // int min_tiles,
strength_floor, //
strength_pow, // double strength_pow,
// update !
smplMode,
smplSide,
smplNum,
smplRms,
dl);
// }
}
}
if (st_planes.size() > 0){
// sort planes by increasing disparity (tile center or plane center ? ) Using plane center
Collections.sort(st_planes, new Comparator<TilePlanes.PlaneData>() {
@Override
public int compare(TilePlanes.PlaneData lhs, TilePlanes.PlaneData rhs) {
// -1 - less than, 1 - greater than, 0 - equal
return (rhs.getZxy()[0] > lhs.getZxy()[0]) ? -1 : (rhs.getZxy()[0] < lhs.getZxy()[0] ) ? 1 : 0;
}
});
if (LOWEST_PLANE(2) > 0) st_planes.add(0, st_planes.get(0)); // insert dummy at pos 0;
result_planes[nsTile] = st_planes.toArray(new TilePlanes.PlaneData[0] );
if (LOWEST_PLANE(2) > 0) result_planes[nsTile][0] = null; // remove dummy
if (dl >0){
System.out.println("createPlanesFromSelections(): nsTile="+nsTile);
}
}
if (dl > 2) {
String [] dbg_titles = showSupertileSeparationTitles( disp_strength[nsTile], plane_selections[nsTile], result_planes[nsTile]);
double [][] dbg_img = showSupertileSeparation(false, disp_strength[nsTile], plane_selections[nsTile], result_planes[nsTile]);
showDoubleFloatArrays sdfa_instance = new showDoubleFloatArrays();
sdfa_instance.showArrays(dbg_img, 2 * superTileSize, 2* superTileSize, true, "create_planes_disp-"+nsTile+"-"+debugLevel,dbg_titles);
dbg_img = showSupertileSeparation(true, disp_strength[nsTile], plane_selections[nsTile], result_planes[nsTile]);
sdfa_instance.showArrays(dbg_img, 2 * superTileSize, 2* superTileSize, true, "create_planes_world-"+nsTile+"-"+debugLevel,dbg_titles);
}
}
}
}
};
}
ImageDtt.startAndJoin(threads);
return result_planes;
// create a list of usable planes according to the mode
return null;
}
public TilePlanes.PlaneData [][] createPlanesFromSelections(
final boolean [][][][] plane_selections, // = new boolean [nStiles][][][]; // num_tiles
final double [][][][] disp_strength,
......@@ -2284,7 +2027,9 @@ public class SuperTiles{
final int stilesX = (tilesX + superTileSize -1)/superTileSize;
final int stilesY = (tilesY + superTileSize -1)/superTileSize;
final int nStiles = stilesX * stilesY;
final Thread[] threads = ImageDtt.newThreadArray(tileProcessor.threadsMax);
// final Thread[] threads = ImageDtt.newThreadArray(tileProcessor.threadsMax);
final Thread[] threads = ImageDtt.newThreadArray((debugLevel > 1)? 1 : tileProcessor.threadsMax);
final AtomicInteger ai = new AtomicInteger(0);
final TilePlanes.PlaneData [][] result_planes = new TilePlanes.PlaneData[nStiles][];
// this.planes = new TilePlanes.PlaneData[nStiles][];
......@@ -2336,6 +2081,52 @@ public class SuperTiles{
dl); // int debugLevel);
if ((st_planes != null) && (!st_planes.isEmpty())){
if (dl > 0){
System.out.println("======= createPlanesFromSelections(): nsTile="+nsTile+" detecting bridges ==========");
}
boolean [][][] split_sels = pd0.filterBridges(
st_planes, // ArrayList<PlaneData> tilePlanes,
3, // int max_grow, // make configurable?
3, // int max_grow_far,
dl); // int debugLevel)
if (split_sels !=null){
if (dl > -1){
System.out.println("======= createPlanesFromSelections(): nsTile="+nsTile+" removing bridges ==========");
}
if (dl > 2) {
String [] dbg_titles = showSupertileSeparationTitles( disp_strength[nsTile], plane_selections[nsTile], result_planes[nsTile]);
double [][] dbg_img = showSupertileSeparation(false, disp_strength[nsTile], plane_selections[nsTile], result_planes[nsTile]);
showDoubleFloatArrays sdfa_instance = new showDoubleFloatArrays();
sdfa_instance.showArrays(dbg_img, 2 * superTileSize, 2* superTileSize, true, "pre_bridge_disp-"+nsTile+"-"+debugLevel,dbg_titles);
dbg_img = showSupertileSeparation(true, disp_strength[nsTile], plane_selections[nsTile], result_planes[nsTile]);
sdfa_instance.showArrays(dbg_img, 2 * superTileSize, 2* superTileSize, true, "pre_bridge_world-"+nsTile+"-"+debugLevel,dbg_titles);
}
st_planes = pd0.createTilePlanesFromSelections(
"" + nsTile+" bridges ", // String suffix,
split_sels, // boolean [][][] plane_selections, // = new boolean [nStiles][][][]; // num_tiles
disp_strength[nsTile], // double [][][] disp_strength,
plDispNorm, // double dispNorm, // Normalize disparities to the average if above
plMinPoints, // int min_tiles,
plTargetEigen, // double plTargetEigen, // = 0.1; // Remove outliers until main axis eigenvalue (possibly scaled by plDispNorm) gets below
0.0, // double plFractOutliers, // = 0.3; // Maximal fraction of outliers to remove
0, // plMaxOutliers, // int plMaxOutliers, // = 20; // Maximal number of outliers to remove
strength_floor, // double strength_floor,
strength_pow, // double strength_pow,
correct_distortions, // boolean correct_distortions,
smplMode, // boolean smplMode, // = true; // Use sample mode (false - regular tile mode)
smplSide, // int smplSide, // = 2; // Sample size (side of a square)
smplNum, // int smplNum, // = 3; // Number after removing worst
smplRms, // double smplRms, // = 0.1; // Maximal RMS of the remaining tiles in a sample
dl); // int debugLevel);
}
}
if ((st_planes != null) && (!st_planes.isEmpty())){
if (LOWEST_PLANE(2) > 0) st_planes.add(0, st_planes.get(0)); // insert dummy at pos 0;
result_planes[nsTile] = st_planes.toArray(new TilePlanes.PlaneData[0] );
if (LOWEST_PLANE(2) > 0) result_planes[nsTile][0] = null; // remove dummy
......@@ -2465,7 +2256,7 @@ public class SuperTiles{
smplNum, // final int smplNum, // = 3; // Number after removing worst
smplRms, // final double smplRms, // = 0.1; // Maximal RMS of the remaining tiles in a sample
debugLevel+1, // final int debugLevel,
debugLevel + 2, // 1, // final int debugLevel,
dbg_X, // final int dbg_X,
dbg_Y); // final int dbg_Y)
this.planes = new_planes; // save as "measured" (as opposed to "smoothed" by neighbors) planes
......@@ -2488,132 +2279,6 @@ public class SuperTiles{
return wh;
}
// TODO: Obsolete (never used)?
TilePlanes.PlaneData [][] getNeibPlanes(
final int dir, // 0: get from up (N), 1:from NE, ... 7 - from NW
final boolean preferDisparity,
final boolean dbgMerge, // Combine 'other' plane with current
final int dbg_X,
final int dbg_Y
)
{
final int tilesX = tileProcessor.getTilesX();
final int tilesY = tileProcessor.getTilesY();
final int superTileSize = tileProcessor.getSuperTileSize();
final int tileSize = tileProcessor.getTileSize();
final int stilesX = (tilesX + superTileSize -1)/superTileSize;
final int stilesY = (tilesY + superTileSize -1)/superTileSize;
final double [] nan_plane = new double [superTileSize*superTileSize];
for (int i = 0; i < nan_plane.length; i++) nan_plane[i] = Double.NaN;
final int [][] dirsYX = {{-1, 0},{-1,1},{0,1},{1,1},{1,0},{1,-1},{0,-1},{-1,-1}};
TilePlanes.PlaneData [][] neib_planes = new TilePlanes.PlaneData[stilesY * stilesX][];
TilePlanes tpl = null; // new TilePlanes(tileSize,superTileSize, geometryCorrection);
// final int debug_stile = 20 * stilesX + 27;
// final int debug_stile = 17 * stilesX + 27;
final int debug_stile = dbg_Y * stilesX + dbg_X;
// final int debug_stile = -1;
for (int sty0 = 0; sty0 < stilesY; sty0++){
int sty = sty0 + dirsYX[dir][0];
for (int stx0 = 0; stx0 < stilesX; stx0++){
int nsTile0 = sty0 * stilesX + stx0; // result tile
int stx = stx0 + dirsYX[dir][1];
int nsTile = sty * stilesX + stx; // from where to get
if ((sty >= 0) && (sty < stilesY) && (stx >= 0) && (stx < stilesX) && (planes[nsTile] != null)) {
neib_planes[nsTile0] = new TilePlanes.PlaneData[planes[nsTile].length];
for (int np = 0; np < planes[nsTile].length; np++){
GeometryCorrection geometryCorrection = planes[nsTile][np].getGeometryCorrection();
boolean correct_distortions = planes[nsTile][np].getCorrectDistortions();
int [] sTileXY0 = {stx0,sty0};
if (tpl == null) {
tpl = new TilePlanes(tileSize,superTileSize, geometryCorrection);
}
TilePlanes.PlaneData pd = tpl.new PlaneData(
sTileXY0,
tileSize,
superTileSize,
geometryCorrection,
correct_distortions);
/*
TilePlanes.PlaneData pd = planes[nsTile0][np].clone();
pd.setS
*/
if (nsTile0 == debug_stile) {
System.out.println("getNeibPlanes(): nsTile0="+nsTile0+", nsTile="+nsTile+", np = "+np+" dir = "+ dir );
}
TilePlanes.PlaneData other_pd = pd.getPlaneToThis(
planes[nsTile][np], // PlaneData otherPd,
(nsTile0 == debug_stile)? 1:0); // int debugLevel)
int num_this = (planes[nsTile0] == null)? 0:(planes[nsTile0].length);
int best_index = 0;
if (dbgMerge && (num_this > 0)) {
double best_value = Double.NaN;
if (num_this > 1){
// find the best candidate for merge
for (int indx = 1; indx < planes[nsTile0].length; indx++){
if (nsTile0 == debug_stile) {
System.out.println("this with this, same weight:");
TilePlanes.PlaneData this_this_pd = planes[nsTile0][indx].mergePlaneToThis(
planes[nsTile0][indx], // PlaneData otherPd,
1.0, // double scale_other,
1.0, // double starWeightPwr, // Use this power of tile weight when calculating connection cost
true, // boolean ignore_weights,
true, // boolean sum_weights,
preferDisparity,
(nsTile0 == debug_stile)? 1:0); // int debugLevel)
System.out.println("other with other, same weight:");
TilePlanes.PlaneData other_other_pd = other_pd.mergePlaneToThis(
other_pd, // PlaneData otherPd,
1.0, // double scale_other,
1.0, // double starWeightPwr, // Use this power of tile weight when calculating connection cost
true, // boolean ignore_weights,
true, // boolean sum_weights,
preferDisparity,
(nsTile0 == debug_stile)? 1:0); // int debugLevel)
System.out.println("other with this, same weight:");
}
TilePlanes.PlaneData merged_pd = planes[nsTile0][indx].mergePlaneToThis(
other_pd, // PlaneData otherPd,
1.0, // double scale_other,
1.0, // double starWeightPwr, // Use this power of tile weight when calculating connection cost
true, // boolean ignore_weights,
true, // boolean sum_weights,
preferDisparity,
(nsTile0 == debug_stile)? 1:0); // int debugLevel)
if (!(merged_pd.getValue() > best_value)) { // Double.isNaN(best_value) will work too
best_value = merged_pd.getValue();
best_index = indx;
}
}
}
// now merge with weights with the best plane of this supertile
if (nsTile0 == debug_stile) {
System.out.println("this with this, proportional weight, np(other):"+np+" best fit(this):"+best_index);
}
TilePlanes.PlaneData merged_pd = planes[nsTile0][best_index].mergePlaneToThis(
other_pd, // PlaneData otherPd,
1.0, // double scale_other,
1.0, // double starWeightPwr, // Use this power of tile weight when calculating connection cost
false, // boolean ignore_weights,
true, // boolean sum_weights,
preferDisparity,
(nsTile0 == debug_stile)? 1:0); // int debugLevel)
if (merged_pd != null) {
merged_pd.scaleWeight(0.5);
}
neib_planes[nsTile0][np] = merged_pd;
} else {
neib_planes[nsTile0][np] = other_pd;
}
}
} else {
neib_planes[nsTile0] = null;
}
}
}
return neib_planes;
}
public int fillSquares()
{
final int tilesX = tileProcessor.getTilesX();
......@@ -2807,7 +2472,7 @@ public class SuperTiles{
final int stilesX = (tilesX + superTileSize -1)/superTileSize;
final int stilesY = (tilesY + superTileSize -1)/superTileSize;
final int dbgTile = dbg_Y * stilesX + dbg_X;
final TileSurface.TileNeibs tnSurface = tileSurface.new TileNeibs(stilesX, stilesY);
final TileNeibs tnSurface = new TileNeibs(stilesX, stilesY);
int [] rslt = {0,0};
for (int nsTile = 0; nsTile < conflicts.length; nsTile++) if (conflicts[nsTile] != null) {
// conflicts may disappear after being fixed, recheck for null
......@@ -2827,7 +2492,7 @@ public class SuperTiles{
start_dir4, // int start_dir4,
right, // boolean right,
reverse, // boolean reverse,
tnSurface, // TileSurface.TileNeibs tnSurface,
tnSurface, // TileNeibs tnSurface,
conflicts, // int [][][] conflicts,
conflict_stats, // Conflicts conflict_stats, // to be updated after applying resolution
starSteps, // How far to look around when calculationg connection cost
......@@ -2857,7 +2522,7 @@ public class SuperTiles{
int start_dir4,
boolean right,
boolean reverse,
TileSurface.TileNeibs tnSurface,
TileNeibs tnSurface,
int [][][] conflicts,
Conflicts conflict_stats, // to be updated after applying resolution
int starSteps, // How far to look around when calculationg connection cost
......@@ -2979,7 +2644,7 @@ public class SuperTiles{
nsTiles[isTile], // int nsTile0,
replacement_tiles, //HashMap<Integer,Integer> replacement_tiles, //
neibs_prev, // int [][][] replacement_neibs,
tnSurface); // TileSurface.TileNeibs tnSurface)
tnSurface); // TileNeibs tnSurface)
}
if (debugLevel > 1) {
System.out.println("Calculating original conflicts");
......@@ -3040,7 +2705,7 @@ public class SuperTiles{
nsTiles[isTile], // int nsTile0,
replacement_tiles, //HashMap<Integer,Integer> replacement_tiles, //
neibs, // neibs, // int [][][] replacement_neibs,
tnSurface); // TileSurface.TileNeibs tnSurface)
tnSurface); // TileNeibs tnSurface)
}
Conflicts new_conflicts_stats = new Conflicts(
new_conflicts,
......@@ -3118,7 +2783,7 @@ public class SuperTiles{
int nl1,
int nl2,
int dir_mask,
TileSurface.TileNeibs tnSurface,
TileNeibs tnSurface,
int [][][] conflicts,
Conflicts conflict_stats, // to be updated after applying resolution
int starSteps, // How far to look around when calculationg connection cost
......@@ -3220,7 +2885,7 @@ public class SuperTiles{
nsTiles[isTile], // int nsTile0,
replacement_tiles, //HashMap<Integer,Integer> replacement_tiles, //
neibs_prev, // int [][][] replacement_neibs,
tnSurface); // TileSurface.TileNeibs tnSurface)
tnSurface); // TileNeibs tnSurface)
}
if (debugLevel > 1) {
System.out.println("Involved supertiles:");
......@@ -3292,7 +2957,7 @@ public class SuperTiles{
nsTiles[isTile], // int nsTile0,
replacement_tiles, //HashMap<Integer,Integer> replacement_tiles, //
neibs_vars[variant], // neibs, // int [][][] replacement_neibs,
tnSurface); // TileSurface.TileNeibs tnSurface)
tnSurface); // TileNeibs tnSurface)
}
variant_conflicts_stats[variant] = new Conflicts(
variant_conflicts[variant],
......@@ -3414,7 +3079,7 @@ public class SuperTiles{
final int stilesX = (tilesX + superTileSize -1)/superTileSize;
final int stilesY = (tilesY + superTileSize -1)/superTileSize;
final int dbgTile = dbg_Y * stilesX + dbg_X;
final TileSurface.TileNeibs tnSurface = tileSurface.new TileNeibs(stilesX, stilesY);
final TileNeibs tnSurface = new TileNeibs(stilesX, stilesY);
int [] rslt = {0,0};
for (int nsTile = 0; nsTile < conflicts.length; nsTile++) if (conflicts[nsTile] != null) {
// conflicts may disappear after being fixed, recheck for null
......@@ -3473,7 +3138,7 @@ public class SuperTiles{
final int stilesX = (tilesX + superTileSize -1)/superTileSize;
final int stilesY = (tilesY + superTileSize -1)/superTileSize;
final int dbgTile = dbg_Y * stilesX + dbg_X;
final TileSurface.TileNeibs tnSurface = tileSurface.new TileNeibs(stilesX, stilesY);
final TileNeibs tnSurface = new TileNeibs(stilesX, stilesY);
int [] rslt = {0,0};
for (int nsTile = 0; nsTile < conflicts.length; nsTile++) if (conflicts[nsTile] != null) {
// conflicts may disappear after being fixed, recheck for null
......@@ -3511,7 +3176,7 @@ public class SuperTiles{
int nsTile,
int nl1,
int nl2,
TileSurface.TileNeibs tnSurface,
TileNeibs tnSurface,
int [][][] conflicts,
Conflicts conflict_stats, // to be updated after applying resolution
int starSteps, // How far to look around when calculationg connection cost
......@@ -3619,7 +3284,7 @@ public class SuperTiles{
nsTiles[isTile], // int nsTile0,
replacement_tiles, //HashMap<Integer,Integer> replacement_tiles, //
neibs_prev, // int [][][] replacement_neibs,
tnSurface); // TileSurface.TileNeibs tnSurface)
tnSurface); // TileNeibs tnSurface)
}
if (debugLevel > 1) {
System.out.println("Involved supertiles:");
......@@ -3649,7 +3314,7 @@ public class SuperTiles{
nsTiles[isTile], // int nsTile0,
replacement_tiles, //HashMap<Integer,Integer> replacement_tiles, //
neibs_vars[variant], // neibs, // int [][][] replacement_neibs,
tnSurface); // TileSurface.TileNeibs tnSurface)
tnSurface); // TileNeibs tnSurface)
}
variant_conflicts_stats[variant] = new Conflicts(
variant_conflicts[variant],
......@@ -3809,7 +3474,7 @@ public class SuperTiles{
int nl1,
int nl2,
LinkPlanes lp,
TileSurface.TileNeibs tnSurface,
TileNeibs tnSurface,
int [][][] conflicts,
Conflicts conflict_stats, // to be updated after applying resolution
int starSteps, // How far to look around when calculationg connection cost
......@@ -3922,7 +3587,7 @@ public class SuperTiles{
nsTiles[isTile], // int nsTile0,
replacement_tiles, // HashMap<Integer,Integer> replacement_tiles, //
neibs_prev, // int [][][] replacement_neibs,
tnSurface); // TileSurface.TileNeibs tnSurface)
tnSurface); // TileNeibs tnSurface)
// calculate cost of conflicts
conflicts_old_cost += iconflicts.getConflictsCost(
......@@ -3932,7 +3597,7 @@ public class SuperTiles{
replacement_tiles, // HashMap<Integer,Integer> replacement_tiles, // null is OK
neibs_prev, // int [][][] replacement_neibs, // null OK if replacement_tiles == null
connectionCosts, // ConnectionCosts connectionCosts,
tnSurface); // TileSurface.TileNeibs tnSurface)
tnSurface); // TileNeibs tnSurface)
}
......@@ -3971,7 +3636,7 @@ public class SuperTiles{
nsTiles[isTile], // int nsTile0,
replacement_tiles, //HashMap<Integer,Integer> replacement_tiles, //
neibs_vars[variant], // neibs, // int [][][] replacement_neibs,
tnSurface); // TileSurface.TileNeibs tnSurface)
tnSurface); // TileNeibs tnSurface)
// connectionCosts now contains last calculated val/weight pairs for broader array of tile data
conflicts_var_cost[variant] += iconflicts.getConflictsCost(
nsTiles[isTile], // int nsTile0,
......@@ -3980,7 +3645,7 @@ public class SuperTiles{
replacement_tiles, // HashMap<Integer,Integer> replacement_tiles, // null is OK
neibs_vars[variant], // int [][][] replacement_neibs, // null OK if replacement_tiles == null
connectionCosts, // ConnectionCosts connectionCosts,
tnSurface); // TileSurface.TileNeibs tnSurface)
tnSurface); // TileNeibs tnSurface)
}
variant_conflicts_stats[variant] = new Conflicts(
variant_conflicts[variant],
......@@ -4159,7 +3824,7 @@ public class SuperTiles{
final int stilesX = (tilesX + superTileSize -1)/superTileSize;
final int stilesY = (tilesY + superTileSize -1)/superTileSize;
final int nStiles = stilesX * stilesY;
final TileSurface.TileNeibs tnSurface = tileSurface.new TileNeibs(stilesX, stilesY);
final TileNeibs tnSurface = new TileNeibs(stilesX, stilesY);
final Thread[] threads = ImageDtt.newThreadArray(tileProcessor.threadsMax);
final AtomicInteger ai = new AtomicInteger(0);
for (int ithread = 0; ithread < threads.length; ithread++) {
......@@ -4174,7 +3839,7 @@ public class SuperTiles{
starValPwr, //double starValPwr, // Raise value of each tile before averaging
steps, // int steps,
planes, // TilePlanes.PlaneData [][] planes,
tnSurface, // TileSurface.TileNeibs tnSurface,
tnSurface, // TileNeibs tnSurface,
preferDisparity); // boolean preferDisparity)
int [] mod_supertiles = new int[1];
for (int nsTile = ai.getAndIncrement(); nsTile < nStiles; nsTile = ai.getAndIncrement()) {
......@@ -4214,7 +3879,7 @@ public class SuperTiles{
// final int tileSize = tileProcessor.getTileSize();
final int stilesX = (tilesX + superTileSize -1)/superTileSize;
final int stilesY = (tilesY + superTileSize -1)/superTileSize;
final TileSurface.TileNeibs tnSurface = tileSurface.new TileNeibs(stilesX, stilesY);
final TileNeibs tnSurface = new TileNeibs(stilesX, stilesY);
final Thread[] threads = ImageDtt.newThreadArray(tileProcessor.threadsMax);
final AtomicInteger ai = new AtomicInteger(0);
for (int ithread = 0; ithread < threads.length; ithread++) {
......@@ -4229,7 +3894,7 @@ public class SuperTiles{
starValPwr, //double starValPwr, // Raise value of each tile before averaging
steps, // int steps,
planes, // TilePlanes.PlaneData [][] planes,
tnSurface, // TileSurface.TileNeibs tnSurface,
tnSurface, // TileNeibs tnSurface,
preferDisparity); // boolean preferDisparity)
int [] supertiles = new int[1];
for (int isTile = ai.getAndIncrement(); isTile < mod_supertiles.length; isTile = ai.getAndIncrement()) {
......@@ -4273,7 +3938,7 @@ public class SuperTiles{
int nl1,
int nl2,
int dir_mask,
TileSurface.TileNeibs tnSurface,
TileNeibs tnSurface,
int debugLevel)
{
int [][] layers0 = new int[5][2]; // 4 - center, 0 - N, 1 - E, 2 - S, 3 - w. 4 has 2, each other 0,1 or 2 layers
......@@ -4684,7 +4349,7 @@ public class SuperTiles{
final int stilesX = (tilesX + superTileSize -1)/superTileSize;
final int stilesY = (tilesY + superTileSize -1)/superTileSize;
final int dbgTile = dbg_Y * stilesX + dbg_X;
final TileSurface.TileNeibs tnSurface = tileSurface.new TileNeibs(stilesX, stilesY);
final TileNeibs tnSurface = new TileNeibs(stilesX, stilesY);
final Conflicts iconflicts = new Conflicts(this);
int [] rslt = {0,0};
......@@ -4870,12 +4535,12 @@ public class SuperTiles{
nsTiles[isTile], // int nsTile0,
replacement_tiles, //HashMap<Integer,Integer> replacement_tiles, //
neibs_prev, // int [][][] replacement_neibs,
tnSurface); // TileSurface.TileNeibs tnSurface)
tnSurface); // TileNeibs tnSurface)
conflicts_new[isTile] = iconflicts.detectTriangularTileConflicts(
nsTiles[isTile], // int nsTile0,
replacement_tiles, //HashMap<Integer,Integer> replacement_tiles, //
neibs, // int [][][] replacement_neibs,
tnSurface); // TileSurface.TileNeibs tnSurface)
tnSurface); // TileNeibs tnSurface)
}
if (debugLevel > 1) {
System.out.println("Involved supertiles:");
......@@ -4974,7 +4639,7 @@ public class SuperTiles{
int dir4,
double maxEigen, // maximal eigenvalue of planes to consider
double starWeightPwr, // Use this power of tile weight when calculating connection cost
TileSurface.TileNeibs tnSurface,
TileNeibs tnSurface,
boolean preferDisparity,
int debugLevel)
{
......@@ -5067,7 +4732,7 @@ public class SuperTiles{
double diagonalWeight,
double starPwr, // Divide cost by number of connections to this power
double starWeightPwr, // Use this power of tile weight when calculating connection cost
TileSurface.TileNeibs tnSurface,
TileNeibs tnSurface,
boolean preferDisparity,
int debugLevel)
{
......@@ -5106,7 +4771,7 @@ public class SuperTiles{
public int [] getInvolvedSupertiles(
int [] mod_supertiles,
TileSurface.TileNeibs tnSurface)
TileNeibs tnSurface)
{
HashSet<Integer> stiles_set = new HashSet<Integer>();
for (int i = 0; i < mod_supertiles.length; i++){
......@@ -5157,7 +4822,7 @@ public class SuperTiles{
double diagonalWeight,
double starPwr, // Divide cost by number of connections to this power
double starWeightPwr, // Use this power of tile weight when calculating connection cost
TileSurface.TileNeibs tnSurface,
TileNeibs tnSurface,
boolean preferDisparity,
int debugLevel)
{
......@@ -6156,7 +5821,7 @@ public class SuperTiles{
smplSide,
smplNum,
smplRms,
dl); // int debugLevel)
dl - 2); // int debugLevel)
if (disp_strength == null) {
System.out.println("=== BUG in applyMergePlanes(): failed to getPlaneFromMeas() for merged planes");
break;
......@@ -6177,7 +5842,7 @@ public class SuperTiles{
smplSide,
smplNum,
smplRms,
dl); // int debugLevel)
dl - 2); // int debugLevel)
......
src/main/java/TileNeibs.java 0 → 100644
View file @ 0ab422e2
import java.util.ArrayList;
import java.util.Arrays;
/**
** TileNeibs - handles walking inside rectangular area
**
** Copyright (C) 2017 Elphel, Inc.
**
** -----------------------------------------------------------------------------**
**
** TileNeibs.java is free software: you can redistribute it and/or modify
** it under the terms of the GNU General Public License as published by
** the Free Software Foundation, either version 3 of the License, or
** (at your option) any later version.
**
** This program is distributed in the hope that it will be useful,
** but WITHOUT ANY WARRANTY; without even the implied warranty of
** MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
** GNU General Public License for more details.
**
** You should have received a copy of the GNU General Public License
** along with this program. If not, see <http://www.gnu.org/licenses/>.
** -----------------------------------------------------------------------------**
**
*/
public class TileNeibs{
int sizeX;
int sizeY;
public int dirs = 8;
public TileNeibs(int size){
this.sizeX = size;
this.sizeY = size;
}
public TileNeibs(int sizeX, int sizeY){
this.sizeX = sizeX;
this.sizeY = sizeY;
}
public int numNeibs() // TODO: make configurable to
{
return dirs;
}
public int opposite(int dir){
return (dir + dirs / 2) % dirs;
}
/**
* Get x,y pair from index
* @param indx element index
* @return array of {x,y}
*/
int [] getXY(int indx)
{
int [] xy = {indx % sizeX ,indx / sizeX};
return xy;
}
/**
* Get element index from x and y
* @param x horizontal position
* @param y vertical position
* @return element linescan index
*/
int getIndex(int x, int y){
return y * sizeX + x;
}
/**
* Get 2d element index after step N, NE, ... NW. Returns -1 if leaving array
* @param indx start index
* @param dir step direction (CW from up)
* @return new index or -1 if leaving
*/
int getNeibIndex(int indx, int dir)
{
int y = indx / sizeX;
int x = indx % sizeX;
if (dir < 0) return indx;
switch (dir % dirs){
case 0: return (y == 0) ? -1 : (indx - sizeX);
case 1: return ((y == 0) || ( x == (sizeX - 1))) ? -1 : (indx - sizeX + 1);
case 2: return ( ( x == (sizeX - 1))) ? -1 : (indx + 1);
case 3: return ((y == (sizeY - 1)) || ( x == (sizeX - 1))) ? -1 : (indx + sizeX + 1);
case 4: return ((y == (sizeY - 1)) ) ? -1 : (indx + sizeX);
case 5: return ((y == (sizeY - 1)) || ( x == 0)) ? -1 : (indx + sizeX - 1);
case 6: return ( ( x == 0)) ? -1 : (indx - 1);
case 7: return ((y == 0) || ( x == 0)) ? -1 : (indx - sizeX - 1);
default: return indx;
}
}
/**
* Return tile segment for 50% overlap. -1 - center, 0 N, 1 - NE,... 7 - NW
* @param indx element index
* @return which of the 9 areas this element belongs
*/
int getSegment(int indx)
{
int s1x = sizeX / 4;
int s3x = 3 * sizeX / 4;
int s1y = sizeY / 4;
int s3y = 3 * sizeY / 4;
int x = indx % sizeX;
int y = indx / sizeX;
boolean up = y < s1y;
boolean down = y >= s3y;
boolean left = x < s1x;
boolean right = x >= s3x;
if (up){
if (left) return 7;
if (right) return 1;
return 0;
}
if (down){
if (left) return 5;
if (right) return 3;
return 4;
}
if (left) return 6;
if (right) return 2;
return -1;
}
/**
* Find if the step leaves the center half of all area
* @param indx start point
* @param dir direction
* @return direction to the new tile (assuming 50% overlap) or -1 if did not cross the border
*/
int leaveOvderlapedCenter(int indx, int dir)
{
int segm = getSegment(indx);
int indx1 = getNeibIndex(indx, dir);
if (indx1 < 0 ) {
return -1; // should not happen
}
int segm1 = getSegment(indx1);
if (segm1 == segm) return -1;
if (segm == -1) return segm1;
int [][] dxy = {
{ 0, 0},
{-1, 0},
{-1, 1},
{ 0, 1},
{ 1, 1},
{ 1, 0},
{ 1,-1},
{ 0,-1},
{-1,-1}};
int dx = dxy[segm1 + 1][1] - dxy[segm + 1][1];
int dy = dxy[segm1 + 1][0] - dxy[segm + 1][0];
for (int dp1 = 0; dp1 <=8; dp1++) {
int sdx = (dx > 0) ? 1: ( (dx < 0) ? -1 : 0);
int sdy = (dy > 0) ? 1: ( (dy < 0) ? -1 : 0);
if ((dxy[dp1][0] == sdy) && (dxy[dp1][1] == sdx)){
return dp1 -1;
}
}
return -1; // should not happen
}
public void growSelection(
int grow, // grow tile selection by 1 over non-background tiles 1: 4 directions, 2 - 8 directions, 3 - 8 by 1, 4 by 1 more
boolean [] tiles,
boolean [] prohibit)
{
boolean [] src_tiles = tiles.clone(); // just in case
// grow
boolean hor = true;
for (; grow > 0; grow--){
boolean single = (grow ==1) && hor;
src_tiles = tiles.clone();
int num_new = 0;
if (hor){
for (int tileY = 0; tileY < sizeY; tileY++){
for (int tileX = 0; tileX < (sizeX - 1); tileX++){
int tindx = tileY * sizeX + tileX;
if ((prohibit == null) || (!prohibit[tindx] && !prohibit[tindx + 1])) {
if (!tiles[tindx + 1] && src_tiles[tindx]){
num_new++;
}
tiles[tindx + 1] |= src_tiles[tindx];
}
}
for (int tileX = 1; tileX < sizeX; tileX++){
int tindx = tileY * sizeX + tileX;
if ((prohibit == null) || (!prohibit[tindx] && !prohibit[tindx - 1])) {
if (!tiles[tindx - 1] && src_tiles[tindx]){
num_new++;
}
tiles[tindx - 1] |= src_tiles[tindx];
}
}
}
}
if (!hor || single){ // do vertically, but from previous state
for (int tileX = 0; tileX < sizeX; tileX++){
for (int tileY = 0; tileY < (sizeY - 1); tileY++){
int tindx = tileY * sizeX + tileX;
if ((prohibit == null) || (!prohibit[tindx] && !prohibit[tindx + sizeX])) {
if (!tiles[tindx + sizeX] && src_tiles[tindx]){
num_new++;
}
tiles[tindx + sizeX] |= src_tiles[tindx];
}
}
for (int tileY = 1; tileY < sizeY; tileY++){
int tindx = tileY * sizeX + tileX;
if ((prohibit == null) || (!prohibit[tindx] && !prohibit[tindx - sizeX])) {
if (!tiles[tindx - sizeX] && src_tiles[tindx]){
num_new++;
}
tiles[tindx - sizeX] |= src_tiles[tindx];
}
}
}
}
hor = !hor;
if (num_new == 0){
break;
}
}
}
public boolean [] boundShape(
boolean [] selection,
boolean octo)
{
boolean [] bound_shape = new boolean [selection.length];
int min_x=-1, max_x=-1, min_y=-1, max_y=-1;
int min_s=-1, max_s=-1, min_d=-1, max_d=-1;
boolean is_set = false;
for (int i = 0; i < selection.length; i++) if (selection[i]){
int [] xy = getXY(i);
int [] sd = {xy[0]+xy[1],xy[0]-xy[1]};
if (!is_set) {
min_x = xy[0];
max_x = xy[0];
min_y = xy[1];
max_y = xy[1];
if (octo) {
min_s = sd[0];
max_s = sd[0];
min_d = sd[1];
max_d = sd[1];
}
is_set = true;
} else {
if (xy[0] < min_x) min_x = xy[0];
else if (xy[0] > max_x) max_x = xy[0];
if (xy[1] < min_y) min_y = xy[1];
else if (xy[1] > max_y) max_y = xy[1];
if (octo) {
if (sd[0] < min_s) min_s = sd[0];
else if (sd[0] > max_s) max_s = sd[0];
if (sd[1] < min_d) min_d = sd[1];
else if (sd[1] > max_d) max_d = sd[1];
}
}
}
for (int y = min_y; y <= max_y; y++){
for (int x = min_x; x <= max_x; x++){
if (!octo ||
(((x + y) >= min_s) && ((x + y) <= max_s) && ((x - y) >= min_d) && ((x - y) <= max_d))) {
bound_shape[getIndex(x,y)] = true;
}
}
}
return bound_shape;
}
/**
* Enumerate clusters on rectangular area
* @param tiles selected tiles, size should be sizeX * sizeY
* @param ordered if true, order tiles from largest to smallest5
* @return integer array, where 0 is unused, 1+ cluster it belongs to
*/
public int [] enumerateClusters(
boolean [] tiles,
boolean ordered)
{
int [] waves = new int [tiles.length];
for (int i = 0; i < tiles.length; i++) waves[i] = tiles[i] ? 0: -1;
ArrayList<Integer> front = new ArrayList<Integer>();
int [] enum_clust = new int[tiles.length];
int numClust = 0;
for (int start_indx = 0; start_indx < tiles.length; start_indx++) if (waves[start_indx] == 0){ // found first pixel of a new cluster
numClust ++;
Integer ipx = start_indx;
Integer ipx1;
front.clear();
int area = 1;
waves[ipx] = area;
enum_clust[ipx] = numClust;
front.add(ipx);
while (!front.isEmpty()) {
ipx = front.remove(0);// get oldest element
for (int d = 0; d < dirs; d++){
ipx1 = getNeibIndex(ipx, d);
if (ipx1 >= 0){
if (waves[ipx1] == 0) {
area++;
waves[ipx1] = area;
enum_clust[ipx1] = numClust;
front.add(ipx1);
}
}
}
}
}
if (!ordered) {
return enum_clust;
}
// count cluster
int []clustSizes = new int [numClust];
for (int i = 0; i < clustSizes.length; i++) clustSizes[i] = 0;
for (int i = 0; i < enum_clust.length; i++) if (enum_clust[i] > 0) clustSizes[enum_clust[i]-1]++;
class Pair implements Comparable<Pair> {
public final int index;
public final int value;
public Pair(int index, int value) {
this.index = index;
this.value = value;
}
@Override
public int compareTo(Pair other) {
return Integer.valueOf(this.value).compareTo(other.value);
}
}
Pair[] pairs = new Pair[numClust];
for (int i = 0; i < clustSizes.length; i++) pairs [i] = new Pair (i, clustSizes[i]);
Arrays.sort(pairs);
int [] revIndex = new int [numClust];
for (int i = 0; i < revIndex.length; i++) revIndex [pairs[i].index] = (numClust - i); // array was in accending order
int [] enum_clust_ordered = new int[tiles.length];
for (int i=0; i < enum_clust_ordered.length; i++){
enum_clust_ordered[i] = (enum_clust[i] > 0) ? revIndex[enum_clust[i] - 1] : 0;
}
return enum_clust_ordered;
}
public int getMax(
int [] data)
{
int mx = data[0];
for (int i = 1; i < data.length; i++){
if (data[i] > mx){
mx= data[i];
}
}
return mx;
}
}
src/main/java/TilePlanes.java
View file @ 0ab422e2
......@@ -22,8 +22,10 @@
**
*/
import java.util.ArrayList;
import java.util.Arrays;
import java.util.Collections;
import java.util.Comparator;
import java.util.HashSet;
import Jama.EigenvalueDecomposition;
import Jama.Matrix;
......@@ -2210,13 +2212,13 @@ public class TilePlanes {
return plane;
}
public double[][] getDoublePlaneDisparityStrength(
public double[][] getDoublePlaneDisparityStrength_old(
double [] window,
boolean use_sel,
boolean divide_by_area,
double scale_projection,
int debugLevel)
{ return getDoublePlaneDisparityStrength(
{ return getDoublePlaneDisparityStrength_old(
true,
window,
use_sel,
......@@ -2259,7 +2261,7 @@ public class TilePlanes {
* @return a pair of arrays {disparity, strength}, each [2*superTileSize * 2*superTileSize]
*/
// obsolete, convert to another version ?
public double[][] getDoublePlaneDisparityStrength(
public double[][] getDoublePlaneDisparityStrength_old(
boolean useWorld,
double [] window,
boolean use_sel,
......@@ -3331,14 +3333,11 @@ public class TilePlanes {
return world_xyz;
}
public ArrayList<TilePlanes.PlaneData> createTilePlanesFromSelections(
public ArrayList<PlaneData> createTilePlanesFromSelections(
String suffix,
boolean [][][] plane_selections, // = new boolean [nStiles][][][]; // num_tiles
double [][][] disp_strength,
// double disp_far, // minimal disparity to select (or NaN)
// double disp_near, // maximal disparity to select (or NaN)
double dispNorm, // Normalize disparities to the average if above
// double min_weight,
int min_tiles,
double plTargetEigen, // = 0.1; // Remove outliers until main axis eigenvalue (possibly scaled by plDispNorm) gets below
double plFractOutliers, // = 0.3; // Maximal fraction of outliers to remove
......@@ -3354,11 +3353,11 @@ public class TilePlanes {
{
// first make a plane from all tiles
ArrayList<TilePlanes.PlaneData> st_planes = new ArrayList<TilePlanes.PlaneData>();
ArrayList<PlaneData> st_planes = new ArrayList<PlaneData>();
// iterate through all plane selections
for (int ps = 0; ps < plane_selections.length; ps++) {
TilePlanes.PlaneData pd = this.clone();
PlaneData pd = this.clone();
boolean OK = (pd.getPlaneFromMeas(
plane_selections[ps], // tile_sel, // boolean [][] tile_sel, // null - do not use, {} use all (will be modified)
disp_strength,
......@@ -3426,7 +3425,6 @@ public class TilePlanes {
}
// calculate the world planes too
// if (debugLevel > -1){
pd.getWorldPlaneFromMeas(
plane_selections[ps], // tile_sel, // boolean [][] tile_sel, // null - do not use, {} use all (will be modified)
disp_strength,
......@@ -3447,9 +3445,9 @@ public class TilePlanes {
}
if (st_planes.size() > 0){
// sort planes by increasing disparity (tile center or plane center ? ) Using plane center
Collections.sort(st_planes, new Comparator<TilePlanes.PlaneData>() {
Collections.sort(st_planes, new Comparator<PlaneData>() {
@Override
public int compare(TilePlanes.PlaneData lhs, TilePlanes.PlaneData rhs) {
public int compare(PlaneData lhs, PlaneData rhs) {
// -1 - less than, 1 - greater than, 0 - equal
return (rhs.getZxy()[0] > lhs.getZxy()[0]) ? -1 : (rhs.getZxy()[0] < lhs.getZxy()[0] ) ? 1 : 0;
}
......@@ -3458,297 +3456,352 @@ public class TilePlanes {
}
return null;
}
}
//TODO: Remove below methods and promote PlaneData (no TilePlanes) after tested
/**
* Calculate covariance matrix for a subset of tile data (disparities)
* Subtract weight floor from weight
* @param data data array - square (2*stSize) * (2*stSize)
* @param weight per sample weight (should have floor already subtracted)
* @param select sample selection
* @return covariance (diagonal) matrix: [0]: disparity, [1]: d<disparity>/dx, [2]: d<disparity>/dy,
*/
public double [][][] getCovar(
double [] data,
double [] weight,
boolean [] select,
double plDispNorm, // Normalize disparities to the average if above
int debugLevel){
double mindet = 1E-15;
int stSize2 = 2 * stSize;
// Matrix covar = new Matrix(3,3);
double [][] acovar = new double [3][3];
int numPoints = 0;
double sw =0.0, swz = 0.0, swx = 0.0, swy = 0.0;
for (int indx = 0; indx < data.length; indx++){
if (select[indx] && (weight[indx] > 0)){
numPoints++;
double w = weight[indx];
double d = data[indx];
// referencing samples to centers of pixels
double x = ((indx % stSize2) - stSize + 0.5) * tileSize + 0.5; // in pixels, not in tiles
double y = ((indx / stSize2) - stSize + 0.5) * tileSize + 0.5;
sw += w;
swz += w * d;
swx += w * x;
swy += w * y;
}
}
if (sw == 0.0) {
return null;
}
swz /= sw;
swx /= sw;
swy /= sw;
// TODO: scale disparity to make same scale for 3 axes?
double kz = ((plDispNorm > 0.0) && (swz > plDispNorm)) ? (plDispNorm / swz) : 1.0;
if (debugLevel > 0){
System.out.println("getCovar(): sw = "+sw +", swz = "+swz +", swx = "+swx +", swy = "+swy+" kz = "+kz);
}
for (int indx = 0; indx < data.length; indx++){
if (select[indx] && (weight[indx] > 0)){
double w = weight[indx] / sw;
double d = kz * (data[indx] - swz);
double wd = w*d;
double x = ((indx % stSize2) - stSize + 0.5) * tileSize + 0.5 - swx;
double y = ((indx / stSize2) - stSize + 0.5) * tileSize + 0.5 - swy;
acovar [0][0] += wd * d;
acovar [0][1] += wd * x;
acovar [0][2] += wd * y;
acovar [1][1] += w * x * x;
acovar [1][2] += w * x * y;
acovar [2][2] += w * y * y;
}
}
acovar [1][0] = acovar [0][1];
acovar [2][0] = acovar [0][2];
acovar [2][1] = acovar [1][2];
Matrix covar = new Matrix(acovar);
// if (Math.abs(covar.det()) < mindet){
// debugLevel = 5;
// }
EigenvalueDecomposition eig = covar.eig();
if (Double.isNaN(eig.getV().get(0, 0))){
System.out.println("getCovar(): Double.isNaN(eig.getV().get(0, 0))");
debugLevel = 20;
}
if (debugLevel > 3){
System.out.println("getCovar(): sw = "+sw +", swz = "+swz +", swx = "+swx +", swy = "+swy +", covar.det() = "+covar.det());
System.out.println("getCovar(): covarianvce matrix, number of used points:"+numPoints);
covar.print(10, 6); // w,d
System.out.println("getCovar(): eigenvalues");
eig.getD().print(10, 6); // w,d
System.out.println("getCovar(): eigenvectors");
eig.getV().print(10, 6); // w,d
}
if ((eig.getD().get(0, 0) == 0.0) || (Math.abs(covar.det()) < mindet)) {
return null; // testing with zero eigenvalue
// Problem with zero eigenvalue is with derivatives and coordinate conversion
}
double [][][] rslt = {
eig.getD().getArray(),
eig.getV().getArray(),
{
{sw,kz,numPoints},
{swz, swx, swy}}};
return rslt;
}
// TODO: obsolete - remove
public PlaneData getPlane(
int [] sTileXY,
double [] data,
double [] weight,
boolean [] select, // null OK, will enable all tiles
boolean correctDistortions,
boolean preferDisparity, // Always start with disparity-most axis (false - lowest eigenvalue)
int debugLevel){
if (select == null) {
select = new boolean [4*stSize];
for (int i = 0; i < select.length; i++) select[i] = true;
}
if (debugLevel > 0){
System.out.println("getPlane()");
}
double [][][] rslt = getCovar(
data,
weight,
select,
0.0,
debugLevel); // debugLevel1); //0); // debugLevel);
if (rslt == null) return null;
int numPoints = (int) rslt[2][0][2];
double swc = rslt[2][0][0];
double [] szxy = rslt[2][1];
double [][] eig_val = rslt[0];
double [][] eig_vect = rslt[1];
// find vector most orthogonal to view // (anyway it all works with that assumption), make it first
// TODO normalize to local linear scales
int oindx = 0;
if (preferDisparity) {
for (int i = 1; i <3; i++){
if (Math.abs(eig_vect[0][i]) > Math.abs(eig_vect[0][oindx])){
oindx = i;
public boolean [][][] refineDiscriminateTiles(
final PlaneData [] planes,
final int stMeasSel, // = 1; // Select measurements for supertiles : +1 - combo, +2 - quad +4 - hor +8 - vert
final double plDispNorm,
final int plMinPoints, // = 5; // Minimal number of points for plane detection
final GeometryCorrection geometryCorrection,
final boolean correct_distortions,
final boolean smplMode, // = true; // Use sample mode (false - regular tile mode)
final int smplSide, // = 2; // Sample size (side of a square)
final int smplNum, // = 3; // Number after removing worst
final double smplRms, // = 0.1; // Maximal RMS of the remaining tiles in a sample
final double max_disp_diff, // maximal disparity difference from the plane to consider tile
final double disp_range, // parallel move known planes around original know value for the best overall fit
final int amplitude_steps, // number of steps (each direction) for each plane to search for the best fit (0 - single, 1 - 1 each side)
final int num_variants, // total number of variants to try (protect from too many planes)
final int mode, // 0 - weighted, 1 - equalized, 2 - best, 3 - combined
final int debugLevel)
{
if (planes == null) return null;
// create a list of usable planes according to the mode
ArrayList<PlaneData> tilePlanes = new ArrayList<PlaneData>();
for (int np = 0; np < planes.length; np++) if (planes[np] != null){
PlaneData weighted_pd = planes[np].getNonexclusiveStar();
PlaneData equal_pd = planes[np].getNonexclusiveStarEq();
switch (mode){
case 0:
if (weighted_pd == null)throw new IllegalArgumentException ("refineDiscriminateTiles(): getNonexclusiveStar() retgurned null");
tilePlanes.add(weighted_pd);
break;
case 1:
if (equal_pd == null)throw new IllegalArgumentException ("refineDiscriminateTiles(): getNonexclusiveStarEq() retgurned null");
tilePlanes.add(equal_pd);
break;
case 2:
if (weighted_pd == null)throw new IllegalArgumentException ("refineDiscriminateTiles(): getNonexclusiveStar() retgurned null");
if (equal_pd == null)throw new IllegalArgumentException ("refineDiscriminateTiles(): getNonexclusiveStarEq() retgurned null");
if (weighted_pd.getWeight() > equal_pd.getWeight()) tilePlanes.add(weighted_pd);
else tilePlanes.add(equal_pd);
break;
case 3:
if (weighted_pd == null)throw new IllegalArgumentException ("refineDiscriminateTiles(): getNonexclusiveStar() retgurned null");
if (equal_pd == null)throw new IllegalArgumentException ("refineDiscriminateTiles(): getNonexclusiveStarEq() retgurned null");
PlaneData combo_pd = weighted_pd.mergePlaneToThis(
equal_pd, // PlaneData otherPd,
1.0, // double scale_other,
1.0, // double starWeightPwr, // Use this power of tile weight when calculating connection cost
false, // boolean ignore_weights,
true, // boolean sum_weights,
preferDisparity, // boolean preferDisparity, // Always start with disparity-most axis (false - lowest eigenvalue)
debugLevel - 2); // int debugLevel)
break;
default:
throw new IllegalArgumentException ("refineDiscriminateTiles(): invalid mode="+mode);
}
}
} else {
for (int i = 1; i < 3 ; i++){
if (eig_val[i][i] < eig_val[oindx][oindx]){
oindx = i;
if (tilePlanes.isEmpty()){
return null;
}
// get measured disparity/strength data, filtered, not tilted
double [][][] disp_strength = new double[measuredLayers.getNumLayers()][][];
for (int ml = 0; ml < disp_strength.length; ml++) if ((stMeasSel & ( 1 << ml)) != 0){
if (smplMode) {
disp_strength[ml] = measuredLayers.getDisparityStrength( // expensive to calculate (improve removing outlayers
ml, // int num_layer,
getSTileXY()[0], // int stX,
getSTileXY()[1], // int stY,
null, // boolean [] sel_in, - use all
strength_floor,
measured_strength_pow, //
smplSide, // = 2; // Sample size (side of a square)
smplNum, // = 3; // Number after removing worst
smplRms, // = 0.1; // Maximal RMS of the remaining tiles in a sample
true); // boolean null_if_none)
} else {
disp_strength[ml] = measuredLayers.getDisparityStrength(
ml, // int num_layer,
getSTileXY()[0], // int stX,
getSTileXY()[1], // int stY,
null, // boolean [] sel_in, - use all
strength_floor, // double strength_floor,
measured_strength_pow, // double strength_pow,
true); // boolean null_if_none);
}
}
double [] window = getWindow (2 * superTileSize);
// get all original planes
int num_planes = tilePlanes.size();
double [][][] pds = new double [num_planes][][];
for (int np = 0; np < pds.length; np++){
PlaneData pd = tilePlanes.get(np);
pds[np] = pd.getDoublePlaneDisparityStrength(
false, // boolean useWorld,
null, // double [] window,
-1, // int dir,
false, // boolean use_sel,
false, // boolean divide_by_area,
1.0, // double scale_projection,
0.0, // double fraction_uni,
0); // int debugLevel)
}
// calculate number of the variants for each plane
int extra_vars = (((int) Math.pow(num_variants, 1.0/num_planes)) - 1) / 2; // 0 - single, 1 - 3 (1 each direction), ...
if (extra_vars > amplitude_steps) extra_vars = amplitude_steps; //
int steps = 2 * extra_vars + 1;
double [] disps = new double [steps];
disps[0] = 0.0; // center
for (int i = 0; i < extra_vars; i++){
disps[2 * i + 1] = (disp_range * (i + 1)) / extra_vars;
disps[2 * i + 2] = -disps[2 * i + 1];
}
int [] state = new int [num_planes]; // variants counter
// for (int variant = 0; i < num_variants; variant++ )
while (true){
// evaluate current variant
double [] weights = new double [num_planes];
double [] err2 = new double [num_planes];
boolean
// calculate next variant
boolean all_done = true;
for (int i = 0; i < num_planes; i ++){
if (state[i] < (steps - 1)){
state[i] ++;
for (int j = 0; j < i; j++){
state[j] = 0;
}
all_done = false;
break;
}
}
if (all_done){
break;
}
}
}
if (eig_val[oindx][oindx] == 0.0){
System.out.println("getPlane(): zero eigenvalue!!");
}
/*
// Find two other axis - "mostly X" (horizontal) and "mostly Y" (vertical)
int vindx = (oindx == 0)? 1 : 0;
int hindx = (oindx == 0)? 2 : ((oindx == 1) ? 2 : 1);
if (Math.abs(eig_vect[2][vindx]) < Math.abs(Math.abs(eig_vect[2][hindx]))){
int tmp = vindx;
vindx = hindx;
hindx = tmp;
}
*/
// select 2 other axes for increasing eigenvalues (so v is short axis, h is the long one)
int vindx = (oindx == 0)? 1 : 0;
int hindx = (oindx == 0)? 2 : ((oindx == 1) ? 2 : 1);
if (eig_val[vindx][vindx] > eig_val[hindx][hindx]){
int tmp = vindx;
vindx = hindx;
hindx = tmp;
}
PlaneData pd = new PlaneData(
sTileXY,
this.tileSize,
this.stSize,
this.geometryCorrection,
correctDistortions);
pd.setZxy(szxy);
// pd.setValues(eig_val[oindx][oindx],eig_val[hindx][hindx],eig_val[vindx][vindx]); // eigenvalues [0] - thickness, 2 other to detect skinny (poles)
pd.setValues(eig_val[oindx][oindx],eig_val[vindx][vindx],eig_val[hindx][hindx]); // eigenvalues [0] - thickness, 2 other to detect skinny (poles)
double [][] plane = {
{eig_vect[0][oindx],eig_vect[1][oindx],eig_vect[2][oindx]}, // plane normal to camera
{eig_vect[0][vindx],eig_vect[1][vindx],eig_vect[2][vindx]}, // "horizontal" axis // to detect skinny planes and poles
{eig_vect[0][hindx],eig_vect[1][hindx],eig_vect[2][hindx]}}; // "vertical" axis // to detect skinny planes and poles
/*
// Make normal be towards camera (positive disparity), next vector - positive in X direction (right), last one - in positive Y (down)
for (int v = 0; v <3; v++) {
if (plane[v][v] < 0.0) for (int i = 0; i < 3; i ++) plane[v][i] = -plane[v][i];
}
*/
// Make normal be towards camera (positive disparity), next vector - positive in X direction (right)
for (int v = 0; v < 2; v++) {
if (plane[v][v] < 0.0) for (int i = 0; i < 3; i ++) plane[v][i] = -plane[v][i];
return null;
}
// make direction last vector so px (x) py (.) disp < 0 (left-hand coordinate system)
if (new Matrix(plane).det() > 0){
for (int i = 0; i < 3; i ++) plane[2][i] = -plane[2][i];
}
pd.setVectors (plane);
pd.setNumPoints (numPoints);
pd.setWeight (swc);
pd.setPlaneSelection(select);
return pd;
}
public PlaneData removePlaneOutliers(
PlaneData pd, // already found or null
int [] sTileXY, // may be null if pd is not null
double [] data,
double [] weight,
boolean [] select, // will be modified
boolean correctDistortions,
double targetEigen, // target eigenvalue for primary axis (is disparity-dependent, so is non-constant)
int maxRemoved, // maximal number of tiles to remove (not a constant)
int minLeft, // minimal number of tiles to keep
boolean preferDisparity, // Always start with disparity-most axis (false - lowest eigenvalue)
int debugLevel){
int stSize2 = 2 * stSize;
if (pd == null) {
pd = getPlane(
sTileXY,
data,
weight,
select, // null OK
correctDistortions,
preferDisparity,
debugLevel);
} else if (select != null){
pd.setPlaneSelection(select);
}
if (pd == null){
return null; // zero eigenvalues
}
if (maxRemoved > (pd.getNumPoints() - minLeft)) maxRemoved = pd.getNumPoints() - minLeft;
int numRemoved = 0;
for (; (pd.getValue() > targetEigen) && (numRemoved < maxRemoved); numRemoved++){
if (debugLevel > 2){
System.out.println("removePlaneOutliers("+sTileXY[0]+":"+sTileXY[1]+"): numRemoved = "+numRemoved+" eigenValue = "+pd.getValue()+" target = "+targetEigen);
}
// make a plane and find the worst (largest disparity difference) tile
// z = -(x*Vx + y*Vy)/Vz
double worst_d2 = 0.0;
int worst_index = -1;
double [] v = pd.getVector();
double [] zxy0 = pd.getZxy();
for (int indx = 0; indx < data.length; indx++){
if (select[indx] && (weight[indx] > 0)){
// double w = weight[indx];
double x = ((indx % stSize2) - stSize) - zxy0[1];
double y = ((indx / stSize2) - stSize) - zxy0[2];
double d = data[indx];
d -= zxy0[0];
d += (x * v[1]+y*v[2])/v[0];
double d2 = d*d;
if (d2 > worst_d2){
worst_d2 = d2;
worst_index = indx;
// detect and split/eliminate planes that are use disconnected tiles
// and are not separated by higher disparity planes
// return planes/selections (no remove outliers!)
boolean[][][] filterBridges(
ArrayList<PlaneData> tilePlanes,
int max_grow_these,
int max_grow_far,
int debugLevel)
{
double [][] pds = new double [tilePlanes.size()][];
double [] max_disp = null;
boolean [][] selections = new boolean[pds.length][];
final int tsize = 4 * stSize * stSize;
for (int np = 0; np < pds.length; np++){
PlaneData pd = tilePlanes.get(np);
if (pd != null){
pds[np] = pd.getDoublePlaneDisparityStrength(
false, // boolean useWorld,
null, // double [] window,
-1, // int dir,
false, // boolean use_sel,
false, // boolean divide_by_area,
1.0, // double scale_projection,
0.0, // double fraction_uni,
0)[0]; // int debugLevel)
if (max_disp == null){
max_disp = pds[np].clone();
} else {
for (int i = 0; i < max_disp.length; i++){
max_disp[i] = Math.max(max_disp[i], pds[np][i]);
}
}
boolean [][] ms = pd.getMeasSelection(); // should not be null
// combine selections from all measurement layers
for (int ml = 0; ml < ms.length; ml++) if (ms[ml] != null){
if (selections[np] == null) {
selections[np] = ms[ml].clone();
} else {
for (int i = 0; i < ms[ml].length; i++) {
selections[np][i] |= ms[ml][i];
}
}
}
}
}
if (worst_index < 0) {
System.out.println("This is a BUG in removePlaneOutliers()");
break;
class TileSelections{
int np; // number of original plane
boolean [] mask; // selection mask to apply to measurement layers
TileSelections(int nl, boolean [] mask){
this.np = nl;
this.mask = mask;
}
}
ArrayList<TileSelections> split_planes = new ArrayList<TileSelections>();
TileNeibs tileNeibs = new TileNeibs(2 * stSize, 2 * stSize);
HashSet<Integer> old_planes = new HashSet<Integer>();
for (int np = 0; np < pds.length; np++){
PlaneData pd = tilePlanes.get(np);
if (pd != null){
old_planes.add(np);
// see if all the plane tiles have other planes with higher disparity
boolean all_far = true;
for (int i = 0; i < max_disp.length; i++)if (selections[np][i]){
if (pds[np][i] == max_disp[i]){
all_far = false;
break;
}
}
if (!all_far){
boolean [] grown_sel = selections[np].clone();
tileNeibs.growSelection( //
max_grow_these, // int grow, // grow tile selection by 1 over non-background tiles 1: 4 directions, 2 - 8 directions, 3 - 8 by 1, 4 by 1 more
grown_sel, // boolean [] tiles,
null); // boolean [] prohibit,
int [] clusters = tileNeibs.enumerateClusters(
grown_sel, // boolean [] tiles,
false); // boolean ordered)
int num_clusters = tileNeibs.getMax(
clusters); // int [] data)
if (num_clusters > 1){
boolean [] dbg_grown_sel =grown_sel.clone();
// determine area that should have connections between these clusters, not blocked by known
// tiles with lower disparity
boolean [] bound_octo= tileNeibs.boundShape(
selections[np], // boolean [] selection,
true); // boolean octo)
// now combine all "offending" selections - known tiles belonging to farther planes
boolean [] sel_offend = new boolean[tsize];
for (int np_other = 0; np_other < pds.length; np_other++) if ((selections[np_other] != null) && (np_other != np)){
for (int i = 0; i < tsize; i++) if (selections[np_other][i]){
sel_offend [i] |= (pds[np][i] > pds[np_other][i]);
}
}
// now grow offending selections to fill gaps
boolean [] dbg_sel_offend = sel_offend.clone();
tileNeibs.growSelection( //
max_grow_far, // int grow, // grow tile selection by 1 over non-background tiles 1: 4 directions, 2 - 8 directions, 3 - 8 by 1, 4 by 1 more
sel_offend, // boolean [] tiles,
null); // boolean [] prohibit,
// create prohibited mask - all not in grown selection that are outside bounding octagon or belong to grown offenders
boolean [] prohibit = new boolean [tsize];
for (int i = 0; i < prohibit.length; i++) {
prohibit[i] = !grown_sel[i] && (!bound_octo[i] || sel_offend[i] ); // (pds[np][i] > max_disp[i]);
}
tileNeibs.growSelection( //
stSize, // int grow, // grow tile selection by 1 over non-background tiles 1: 4 directions, 2 - 8 directions, 3 - 8 by 1, 4 by 1 more
grown_sel, // boolean [] tiles,
prohibit); // boolean [] prohibit,
// re-clusterize what is remaining
int [] dbg_clusters = clusters.clone();
clusters = tileNeibs.enumerateClusters(
grown_sel, // boolean [] tiles,
false); // boolean ordered)
num_clusters = tileNeibs.getMax(
clusters);
if (num_clusters > 1){
boolean [][] cluster_sels = new boolean [num_clusters][selections[np].length];
for (int i = 0; i < selections[np].length; i++) if (selections[np][i]){
cluster_sels[clusters[i]-1][i] = true;
}
for (int nc = 0; nc < num_clusters; nc++){
split_planes.add(new TileSelections(np, cluster_sels[nc]));
}
old_planes.remove(np);
}
if (debugLevel > 2) {
String [] dbg_titles = {"sel","grown", "clust0","bound","offend","off_grown", "prohibit", "clusters","clus_masked"};
double [][] dbg_img = new double [dbg_titles.length][tsize];
double lon = 3.0;
for (int i = 0; i < tsize; i++) {
dbg_img[0][i] = selections[np][i] ? lon: 0.0;
dbg_img[1][i] = dbg_grown_sel[i] ? lon: 0.0;
dbg_img[2][i] = dbg_clusters[i];
dbg_img[3][i] = bound_octo[i] ? lon: 0.0;
dbg_img[4][i] = dbg_sel_offend[i] ? lon: 0.0;
dbg_img[5][i] = sel_offend[i] ? lon: 0.0;
dbg_img[6][i] = prohibit[i] ? lon: 0.0;
dbg_img[7][i] = clusters[i];
dbg_img[8][i] = selections[np][i] ? clusters[i] : 0.0;
}
showDoubleFloatArrays sdfa_instance = new showDoubleFloatArrays();
sdfa_instance.showArrays(dbg_img, 2 * superTileSize, 2* superTileSize, true, "bridges-"+np+"-"+debugLevel,dbg_titles);
sdfa_instance.showArrays(pds, 2 * superTileSize, 2* superTileSize, true, "pds-bridges-"+np+"-"+debugLevel,dbg_titles);
}
}
}
}
}
select[worst_index] = false;
if (debugLevel > 2){
System.out.println("removePlaneOutliers() worst_index = " + worst_index);
if (!split_planes.isEmpty()){
boolean [][][] split_selections = new boolean [old_planes.size() + split_planes.size()][][];
int ns = 0;
for (Integer np: old_planes){
split_selections[ns++] = tilePlanes.get(np).getMeasSelection();
}
for (TileSelections ts: split_planes){
PlaneData pd = tilePlanes.get(ts.np);
boolean [][] ms = tilePlanes.get(ts.np).getMeasSelection().clone();
for (int ml = 0; ml < ms.length; ml++) if (ms[ml] != null){
ms[ml] = ms[ml].clone();
for (int i = 0; i < ms[ml].length; i++){
ms[ml][i] &= ts.mask[i];
}
}
split_selections[ns++] = ms;
}
return split_selections;
}
pd = getPlane( // re-calculate with point removed
pd.getSTileXY(),
data,
weight,
select,
correctDistortions,
preferDisparity,
debugLevel);
if (pd == null) {
return null;
return null; // no changes
}
public double [] getWindow (
int size)
{
double [] wnd1d = new double [size];
for (int i = 0; i < size/2; i++){
wnd1d[i] = 0.5 * (1.0 - Math.cos(2*Math.PI*(i+0.5)/size));
wnd1d[size - i -1] = wnd1d[i];
}
double [] wnd = new double [size * size];
int indx = 0;
for (int i = 0; i < size; i++){
for (int j = 0; j < size; j++){
wnd[indx++] = wnd1d[i]*wnd1d[j];
}
}
return wnd;
}
return pd;
}
}
src/main/java/TileProcessor.java
View file @ 0ab422e2
......@@ -3386,7 +3386,7 @@ public class TileProcessor {
LinkPlanes lp = new LinkPlanes (clt_parameters, st);
// try to merge multiple times
int max_num_merge_try = 5;
int max_num_merge_try = 8;
TilePlanes.PlaneData [][][] dbg_orig_planes = new TilePlanes.PlaneData [max_num_merge_try][][];
for (int num_merge_try = 0; num_merge_try < max_num_merge_try; num_merge_try++){
......@@ -3467,20 +3467,6 @@ public class TileProcessor {
clt_parameters.tileX,
clt_parameters.tileY);
// System.out.println("merge_cost_data.length = " + merge_cost_data.length);
/*
// double [][][][][][] merge_cost_data_eq =
lp.costSameTileConnections(
true, // final boolean ignore_weights,
1.8, // final double threshold_worst,
1000.0, //final double threshold_world_worst,
st.planes, // ffinal TilePlanes.PlaneData [][] planes,
merge_candidates, // final int [][][] merge_candidates,
plane_nooverlaps, // final boolean [][][] valid_candidates, // will be updated
2, // -1, // debugLevel, // final int debugLevel)
clt_parameters.tileX,
clt_parameters.tileY);
// System.out.println("merge_cost_data_eq.length = " + merge_cost_data_eq.length);
*/
int [][][] merge_groups = lp.extractMergeSameTileGroups(
st.planes, // final TilePlanes.PlaneData [][] planes,
merge_candidates, // final int [][][] merge_candidates,
......
src/main/java/TileSurface.java
View file @ 0ab422e2
......@@ -235,144 +235,6 @@ public class TileSurface {
}
public class TileNeibs{
int sizeX;
int sizeY;
public int dirs = 8;
public TileNeibs(int size){
this.sizeX = size;
this.sizeY = size;
}
public TileNeibs(int sizeX, int sizeY){
this.sizeX = sizeX;
this.sizeY = sizeY;
}
public int numNeibs() // TODO: make configurable to
{
return dirs;
}
public int opposite(int dir){
return (dir + dirs / 2) % dirs;
}
/**
* Get x,y pair from index
* @param indx element index
* @return array of {x,y}
*/
int [] getXY(int indx)
{
int [] xy = {indx % sizeX ,indx / sizeX};
return xy;
}
/**
* Get element index from x and y
* @param x horizontal position
* @param y vertical position
* @return element linescan index
*/
int getIndex(int x, int y){
return y * sizeX + x;
}
/**
* Get 2d element index after step N, NE, ... NW. Returns -1 if leaving array
* @param indx start index
* @param dir step direction (CW from up)
* @return new index or -1 if leaving
*/
int getNeibIndex(int indx, int dir)
{
int y = indx / sizeX;
int x = indx % sizeX;
if (dir < 0) return indx;
switch (dir % dirs){
case 0: return (y == 0) ? -1 : (indx - sizeX);
case 1: return ((y == 0) || ( x == (sizeX - 1))) ? -1 : (indx - sizeX + 1);
case 2: return ( ( x == (sizeX - 1))) ? -1 : (indx + 1);
case 3: return ((y == (sizeY - 1)) || ( x == (sizeX - 1))) ? -1 : (indx + sizeX + 1);
case 4: return ((y == (sizeY - 1)) ) ? -1 : (indx + sizeX);
case 5: return ((y == (sizeY - 1)) || ( x == 0)) ? -1 : (indx + sizeX - 1);
case 6: return ( ( x == 0)) ? -1 : (indx - 1);
case 7: return ((y == 0) || ( x == 0)) ? -1 : (indx - sizeX - 1);
default: return indx;
}
}
/**
* Return tile segment for 50% overlap. -1 - center, 0 N, 1 - NE,... 7 - NW
* @param indx element index
* @return which of the 9 areas this element belongs
*/
int getSegment(int indx)
{
int s1x = sizeX / 4;
int s3x = 3 * sizeX / 4;
int s1y = sizeY / 4;
int s3y = 3 * sizeY / 4;
int x = indx % sizeX;
int y = indx / sizeX;
boolean up = y < s1y;
boolean down = y >= s3y;
boolean left = x < s1x;
boolean right = x >= s3x;
if (up){
if (left) return 7;
if (right) return 1;
return 0;
}
if (down){
if (left) return 5;
if (right) return 3;
return 4;
}
if (left) return 6;
if (right) return 2;
return -1;
}
/**
* Find if the step leaves the center half of all area
* @param indx start point
* @param dir direction
* @return direction to the new tile (assuming 50% overlap) or -1 if did not cross the border
*/
int leaveOvderlapedCenter(int indx, int dir)
{
int segm = getSegment(indx);
int indx1 = getNeibIndex(indx, dir);
if (indx1 < 0 ) {
return -1; // should not happen
}
int segm1 = getSegment(indx1);
if (segm1 == segm) return -1;
if (segm == -1) return segm1;
int [][] dxy = {
{ 0, 0},
{-1, 0},
{-1, 1},
{ 0, 1},
{ 1, 1},
{ 1, 0},
{ 1,-1},
{ 0,-1},
{-1,-1}};
int dx = dxy[segm1 + 1][1] - dxy[segm + 1][1];
int dy = dxy[segm1 + 1][0] - dxy[segm + 1][0];
for (int dp1 = 0; dp1 <=8; dp1++) {
int sdx = (dx > 0) ? 1: ( (dx < 0) ? -1 : 0);
int sdy = (dy > 0) ? 1: ( (dy < 0) ? -1 : 0);
if ((dxy[dp1][0] == sdy) && (dxy[dp1][1] == sdx)){
return dp1 -1;
}
}
return -1; // should not happen
}
}
public int getNStileDir(
int nsTile,
int dir)
......
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