Debugging field calibration with GPU

src/main/java/com/elphel/imagej/cameras/CLTParameters.java

@@ -167,7 +167,7 @@ public class CLTParameters {
 	public boolean    inf_restore_disp = true;   // Add disparity back to d{x,y}[i] (debug feature)
 	// Lazy eye parameters
 
-	public boolean    ly_lma_ers =      true;    // Use 2020 LMA-based measurement of mismatch
+	public boolean    ly_lma_ers =      true;    // Use 2020 LMA-based measurement of mismatch (GPU-supported)
 	public double     ly_gt_strength =  0.18;    // use some configurable parameters
 	public boolean    ly_gt_use_wnd =   true;    //
 	public double     ly_gt_rms =       0.2;     // split small source samples to FG/BG if all aux tile RMS exceeds this value
@@ -797,6 +797,7 @@ public class CLTParameters {
 	public boolean    gpu_use_aux =         false; // accelerate tile processor for the aux (lwir) quad camera
 	public boolean    gpu_use_aux_macro =   false; // accelerate tile processor for the aux (lwir) quad camera in macro mode
 	public boolean    gpu_use_aux_adjust =  false; // accelerate tile processor for the aux (lwir) quad camera for field calibration
+	public boolean    gpu_debug_accum =     false; // debug multi-tile TD accumulation
 
 	public boolean    replaceWeakOutliers =   true; // false;
 
@@ -1600,6 +1601,7 @@ public class CLTParameters {
 		properties.setProperty(prefix+"gpu_use_aux",                this.gpu_use_aux +"");
 		properties.setProperty(prefix+"gpu_use_aux_macro",          this.gpu_use_aux_macro +"");
 		properties.setProperty(prefix+"gpu_use_aux_adjust",         this.gpu_use_aux_adjust +"");
+		properties.setProperty(prefix+"gpu_debug_accum",            this.gpu_debug_accum +"");
 		
 		properties.setProperty(prefix+"debug_initial_discriminate",       this.debug_initial_discriminate+"");
 		properties.setProperty(prefix+"dbg_migrate",                      this.dbg_migrate+"");
@@ -2387,6 +2389,7 @@ public class CLTParameters {
 		if (properties.getProperty(prefix+"gpu_use_aux")!=null)                 this.gpu_use_aux=Boolean.parseBoolean(properties.getProperty(prefix+"gpu_use_aux"));
 		if (properties.getProperty(prefix+"gpu_use_aux_macro")!=null)           this.gpu_use_aux_macro=Boolean.parseBoolean(properties.getProperty(prefix+"gpu_use_aux_macro"));
 		if (properties.getProperty(prefix+"gpu_use_aux_adjust")!=null)          this.gpu_use_aux_adjust=Boolean.parseBoolean(properties.getProperty(prefix+"gpu_use_aux_adjust"));
+		if (properties.getProperty(prefix+"gpu_debug_accum")!=null)             this.gpu_debug_accum=Boolean.parseBoolean(properties.getProperty(prefix+"gpu_debug_accum"));
 
 		if (properties.getProperty(prefix+"debug_initial_discriminate")!=null)       this.debug_initial_discriminate=Boolean.parseBoolean(properties.getProperty(prefix+"debug_initial_discriminate"));
 		if (properties.getProperty(prefix+"dbg_migrate")!=null)                      this.dbg_migrate=Boolean.parseBoolean(properties.getProperty(prefix+"dbg_migrate"));
@@ -2612,7 +2615,7 @@ public class CLTParameters {
 
 		gd.addTab         ("Lazy eye", "Lazy eye parameters");
 
-		gd.addCheckbox    ("Use 2020 LMA-based measurement of mismatch",                                        this.ly_lma_ers);
+		gd.addCheckbox    ("Use 2020 LMA-based measurement of mismatch (GPU-supported)",                        this.ly_lma_ers);
 		gd.addMessage     ("--- main-to-aux depth map parameters ---");
 		gd.addNumericField("Minimal reference (main) channel correlation strength",                             this.ly_gt_strength,        3);
 		gd.addCheckbox    ("Use window for AUX tiles to reduce weight of the hi-res tiles near low-res tile boundaries", this.ly_gt_use_wnd);
@@ -3337,6 +3340,7 @@ public class CLTParameters {
 		gd.addCheckbox    ("Accelerate tile processor for the aux (lwir) quad camera",                                  this.gpu_use_aux);
 		gd.addCheckbox    ("Accelerate tile processor for the aux (lwir) quad camera in macro mode",                    this.gpu_use_aux_macro);
 		gd.addCheckbox    ("Accelerate tile processor for the aux (lwir) quad camera for field calibration",            this.gpu_use_aux_adjust);
+		gd.addCheckbox    ("Debug multi-tile TD accumulation",                                                          this.gpu_debug_accum);
 		
 		gd.addTab         ("LWIR", "parameters for LWIR/EO 8-camera rig");
 		this.lwir.dialogQuestions(gd);
@@ -4097,6 +4101,7 @@ public class CLTParameters {
 		this.gpu_use_aux=           gd.getNextBoolean();
 		this.gpu_use_aux_macro=     gd.getNextBoolean();
 		this.gpu_use_aux_adjust=    gd.getNextBoolean();
+		this.gpu_debug_accum=       gd.getNextBoolean();
 		
 		this.lwir.dialogAnswers(gd);
 

src/main/java/com/elphel/imagej/gpu/GPUTileProcessor.java

@@ -222,6 +222,18 @@ public class GPUTileProcessor {
     			}
     		}
     	}
+    	public float [][] getDispDist(){
+    		return disp_dist;
+    	}
+    	public float [][] getXY(boolean use_aux){
+    		return use_aux? xy_aux : xy;
+    	}
+    	public int getTileY(){
+    		return ty;
+    	}
+    	public int getTileX(){
+    		return tx;
+    	}
 
     	// convert this class instance to float array to match layout of the C struct
     	public float [] asFloatArray(boolean use_aux) {
@@ -2043,10 +2055,13 @@ public class GPUTileProcessor {
             	}
         	}
         	setCorrIndicesTdData(
-        			ntile,   // int    num_tiles,  // corr_indices, fdata may be longer than needed
+        			ntile * num_pairs,   // int    num_tiles,  // corr_indices, fdata may be longer than needed
         			indices, // int [] corr_indices,
             		fdata);  // float [] fdata);
-        	return indices;
+        	// trim indices
+        	int [] indices_trim = new int [ntile * num_pairs];
+        	System.arraycopy(indices, 0, indices_trim, 0 , indices_trim.length);        	
+        	return indices_trim;
         }
         
         public float [][][][] getCorrTilesTd() // [tileY][tileX][pair][4*64] , read all available pairs
@@ -2104,7 +2119,9 @@ public class GPUTileProcessor {
         			ntile,   // int    num_tiles,  // corr_indices, fdata may be longer than needed
         			indices, // int [] corr_indices,
             		fdata);  // float [] fdata);
-        	return indices;
+        	int [] indices_trim = new int [ntile];
+        	System.arraycopy(indices, 0, indices_trim, 0 , indices_trim.length);        	
+        	return indices_trim;
         }
         
         public float [][][] getCorrTilesComboTd() // [tileY][tileX][4*64] , read all available pairs

src/main/java/com/elphel/imagej/tileprocessor/ImageDttCPU.java

@@ -1509,7 +1509,7 @@ public class ImageDttCPU {
 	}
 
 // removing macro and FPGA modes
-	public double [][] cltMeasureLazyEye ( // returns d,s lazy eye parameters
+	public double[][] cltMeasureLazyEye ( // returns d,s lazy eye parameters
 			final ImageDttParameters  imgdtt_params,   // Now just extra correlation parameters, later will include, most others
 			final int [][]            tile_op,         // [tilesY][tilesX] - what to do - 0 - nothing for this tile
 			final double [][]         disparity_array, // [tilesY][tilesX] - individual per-tile expected disparity
@@ -1753,6 +1753,7 @@ public class ImageDttCPU {
 // filter only tiles with similar disparity to enable lazy eye for the ERS.
 						int num_good_tiles = 0;
 						while (true) {
+//							num_good_tiles = 0; // FIXME: Was missing - uncomment?
 							int mnTx = -1, mnTy = -1, mxTx = -1, mxTy = -1;
 							double mn = Double.NaN;
 							double mx = Double.NaN;
@@ -2148,7 +2149,7 @@ public class ImageDttCPU {
 														imgdtt_params.lmas_max_rel_rms,  // maximal relative (to average max/min amplitude LMA RMS) // May be up to 0.3)
 														imgdtt_params.lmas_min_strength, // minimal composite strength (sqrt(average amp squared over absolute RMS)
 														imgdtt_params.lmas_min_ac,       // minimal of A and C coefficients maximum (measures sharpest point/line)
-														imgdtt_params.lmas_max_area,     //double  lma_max_area,     // maximal half-area (if > 0.0)
+														imgdtt_params.lmas_max_area,     // double  lma_max_area,     // maximal half-area (if > 0.0)
 														imgdtt_params.lma_str_scale,     // convert lma-generated strength to match previous ones - scale
 														imgdtt_params.lma_str_offset     // convert lma-generated strength to match previous ones - add to result
 														)[0];

src/main/java/com/elphel/imagej/tileprocessor/QuadCLTCPU.java

@@ -7403,17 +7403,37 @@ public class QuadCLTCPU {
     					  tp.clt_3d_passes.get(refine_pass), // CLTPass3d   scan,
     					  "after_refinePassSetup-"+tp.clt_3d_passes.size());
 
+    			  // Just debugging
+    			  if (clt_parameters.gpu_debug_accum) {
+    				  CLTMeasureCorrTesting( // perform single pass according to prepared tiles operations and disparity
+    						  clt_parameters,
+    						  0,
+    						  false, // true, // final boolean     save_textures,
+    						  threadsMax,  // maximal number of threads to launch
+    						  updateStatus,
+    						  debugLevel);
+
+    				  CLTMeasureCorrTesting( // perform single pass according to prepared tiles operations and disparity
+    						  clt_parameters,
+    						  refine_pass,
+    						  false, // true, // final boolean     save_textures,
+    						  threadsMax,  // maximal number of threads to launch
+    						  updateStatus,
+    						  debugLevel);
+    			  }
+      			  // End of just debugging
+    			  
     			  CLTMeasureCorr( // perform single pass according to prepared tiles operations and disparity
-//    					  image_data, // first index - number of image in a quad
-//    					  saturation_imp, //final boolean [][]  saturation_imp, // (near) saturated pixels or null
     					  clt_parameters,
     					  refine_pass,
     					  false, // true, // final boolean     save_textures,
     					  threadsMax,  // maximal number of threads to launch
     					  updateStatus,
     					  debugLevel);
-    			  if (debugLevel > -1){
-    				  System.out.println("CLTMeasure("+refine_pass+")");
+    			  
+    			  
+    			  if (debugLevel > -3){
+    				  System.out.println("CLTMeasure("+refine_pass+")-*");
     			  }
     			  if (show_init_refine) tp.showScan(
     					  tp.clt_3d_passes.get(refine_pass), // CLTPass3d   scan,
@@ -7664,12 +7684,28 @@ public class QuadCLTCPU {
 		  for (int  nTile = 0 ; nTile < bg_use.length; nTile++) {
 			  if (bg_sel[nTile] &&
 					  (filtered_bgnd_disp_strength[1][nTile] > 0.0) &&
-					  (bg_str[nTile] > clt_parameters.fcorr_inf_strength) &&
-					  ((bg_overexp == null) || (bg_overexp[nTile] < clt_parameters.lym_overexp))
+					  (bg_str[nTile] > clt_parameters.fcorr_inf_strength) && // 0.13
+					  ((bg_overexp == null) || (bg_overexp[nTile] < clt_parameters.lym_overexp)) //1e-4
 					  ){
 				  bg_use[nTile] = true;
 			  }
 		  }
+		  if (true) {
+			  String [] dbg_titles = {"fdisp", "fstr", "disp", "str", "overexp","sel","use"};
+			  double [][] ddd = {filtered_bgnd_disp_strength[0],filtered_bgnd_disp_strength[1],null,bg_str,bg_overexp, null, null};
+			  ddd[5] = new double [bg_sel.length];
+			  ddd[6] = new double [bg_sel.length];
+			  for (int  nTile = 0 ; nTile < bg_use.length; nTile++) {
+				  ddd[5][nTile] = bg_sel[nTile]?1.0:0.0;
+				  ddd[6][nTile] = bg_use[nTile]?1.0:0.0;
+			  }
+			  (new ShowDoubleFloatArrays()).showArrays(
+				  ddd,
+				  tp.getTilesX(),
+				  tp.getTilesY(),
+				  true,
+				  "filtered_bgnd_disp_strength",dbg_titles);
+		  }
 		  int num_bg = tp.clt_3d_passes.get(bg_scan).setTileOpDisparity( // other minimal strength?
 				  bg_use, // boolean [] selection,
 				  null); // double []  disparity); // null for 0
@@ -7694,12 +7730,8 @@ public class QuadCLTCPU {
 			  System.out.println("Number of background tiles = " + num_bg+", number of lazy eye tiles = " + num_combo);
 		  }
 		  // measure combo
-
-
-
-		  CLTMeasure( // perform single pass according to prepared tiles operations and disparity
-//				  image_data, // first index - number of image in a quad
-//				  saturation_imp, // boolean [][] saturation_imp, // (near) saturated pixels or null
+/*
+		  CLTMeasure( // perform single pass according to prepared tiles operations and disparity **** CPU?
 				  clt_parameters,
 				  combo_scan,
 				  false, // final boolean     save_textures,
@@ -7708,9 +7740,18 @@ public class QuadCLTCPU {
 				  tp.threadsMax,  // maximal number of threads to launch
 				  false, // updateStatus,
 				  debugLevelInner - 1);
-		  if (!batch_mode && clt_parameters.show_extrinsic && (debugLevel >-1)) {
+
+ */
+		  CLTMeasureCorr( // perform single pass according to prepared tiles operations and disparity **** CPU?
+				  clt_parameters,
+				  combo_scan,
+				  false, // final boolean     save_textures,
+				  tp.threadsMax,  // maximal number of threads to launch
+				  false, // updateStatus,
+				  debugLevelInner - 1);
+		  if (!batch_mode && clt_parameters.show_extrinsic && (debugLevel >-3)) {
 			  tp.showScan(
-					  tp.clt_3d_passes.get(bg_scan),   // CLTPass3d   scan,
+					  tp.clt_3d_passes.get(bg_scan),   // CLTPass3d   scan, badly filtered?
 					  "bg_scan_post"); //String title)
 			  tp.showScan(
 					  tp.clt_3d_passes.get(combo_scan),   // CLTPass3d   scan,
@@ -7755,7 +7796,25 @@ public class QuadCLTCPU {
 				  combo_use[nTile] = true;
 			  }
 		  }
-		  int num_combo1 = tp.clt_3d_passes.get(combo_scan).setTileOpDisparity(
+		  if (true) {
+			  String [] dbg_titles = {"fdisp", "fstr", "disp", "str", "overexp","sel","use"};
+			  double [][] ddd = {filtered_combo_scand_isp_strength[0],filtered_combo_scand_isp_strength[1],combo_disp,combo_str,combo_overexp, null, null};
+			  ddd[5] = new double [combo_use.length];
+			  ddd[6] = new double [combo_use.length];
+			  for (int  nTile = 0 ; nTile < combo_use.length; nTile++) {
+				  //ddd[5][nTile] = bg_sel[nTile]?1.0:0.0;
+				  ddd[6][nTile] = combo_use[nTile]?1.0:0.0;
+			  }
+			  (new ShowDoubleFloatArrays()).showArrays(
+					  ddd,
+					  tp.getTilesX(),
+					  tp.getTilesY(),
+					  true,
+					  "filtered_combo_scand_isp_strength",dbg_titles);
+		  }
+		  
+		  
+		  int num_combo1 = tp.clt_3d_passes.get(combo_scan).setTileOpDisparity( // GPU ==0 !
 				  combo_use, // boolean [] selection,
 				  combo_disp); // double []  disparity);
 		  if (debugLevel > -1) {
@@ -7772,7 +7831,7 @@ public class QuadCLTCPU {
 				  dbg_bg_use[i] =    bg_use[i]? 1.0:0.0;
 				  dbg_combo_use[i] = combo_use[i]? 1.0:0.0;
 			  }
-			  double [][]dbg_img = {
+			  double [][]dbg_img = { // bg_use - all 0? (never assigned)?
 					  filtered_bgnd_disp_strength[0],
 					  filtered_bgnd_disp_strength[1],
 					  filtered_combo_scand_isp_strength[0],
@@ -7816,9 +7875,9 @@ public class QuadCLTCPU {
 						  false, // updateStatus,
 						  debugLevelInner -1); // - 1); // -5-1
 				  if (debugLevel > -2) {
-				  tp.showScan(
-						  tp.clt_3d_passes.get(combo_scan),   // CLTPass3d   scan,
-						  "LY_combo_scan-"+combo_scan+"_post"); //String title)
+					  tp.showScan(
+							  tp.clt_3d_passes.get(combo_scan),   // CLTPass3d   scan,
+							  "LY_combo_scan-"+combo_scan+"_post"); //String title)
 				  }
 
 				  int tilesX = tp.getTilesX();
@@ -7913,9 +7972,8 @@ public class QuadCLTCPU {
 					  break;
 				  }
 				  if (update_disp_from_latest) {
+/*					  
 					  CLTMeasure( // perform single pass according to prepared tiles operations and disparity
-//							  image_data,       // first index - number of image in a quad
-//							  saturation_imp,   // boolean [][] saturation_imp, // (near) saturated pixels or null
 							  clt_parameters,
 							  combo_scan,
 							  false,             // final boolean     save_textures,
@@ -7924,14 +7982,20 @@ public class QuadCLTCPU {
 							  tp.threadsMax,     // maximal number of threads to launch
 							  false,             // updateStatus,
 							  debugLevelInner - 1);
+*/							  
+					  CLTMeasureCorr( // perform single pass according to prepared tiles operations and disparity
+							  clt_parameters,
+							  combo_scan,
+							  false,             // final boolean     save_textures,
+							  tp.threadsMax,     // maximal number of threads to launch
+							  false,             // updateStatus,
+							  debugLevelInner - 1);
 				  }
 
-			  } else {
+			  } else { // Old, no-GPU
 				  double [][] bg_mismatch = new double[12][];
 				  double [][] combo_mismatch = new double[12][];
 				  CLTMeasure( // perform single pass according to prepared tiles operations and disparity
-//						  image_data, // first index - number of image in a quad
-//						  saturation_imp, // boolean [][] saturation_imp, // (near) saturated pixels or null
 						  clt_parameters,
 						  bg_scan,
 						  false,             // final boolean     save_textures,
@@ -7941,8 +8005,6 @@ public class QuadCLTCPU {
 						  false, // updateStatus,
 						  debugLevelInner - 1);
 				  CLTMeasure( // perform single pass according to prepared tiles operations and disparity
-//						  image_data, // first index - number of image in a quad
-//						  saturation_imp, // boolean [][] saturation_imp, // (near) saturated pixels or null
 						  clt_parameters,
 						  combo_scan,
 						  false,             // final boolean     save_textures,
@@ -11317,4 +11379,21 @@ public class QuadCLTCPU {
 		  }
 		  scan.setTilesRBGA(tileTones); 
 	  }
+	  // non-trivial in QuadCLT (for the GPU)
+	  public CLTPass3d  CLTMeasureCorrTesting( // perform single pass according to prepared tiles operations and disparity // not used in lwir
+			  CLTParameters     clt_parameters,
+			  final int         scanIndex,
+			  final boolean     save_textures,
+			  final int         threadsMax,  // maximal number of threads to launch
+			  final boolean     updateStatus,
+			  final int         debugLevel)
+	  {
+		 return CLTMeasureCorrTesting( // perform single pass according to prepared tiles operations and disparity // not used in lwir
+				  clt_parameters,
+				  scanIndex,
+				  save_textures,
+				  threadsMax,  // maximal number of threads to launch
+				  updateStatus,
+				  debugLevel); 
+	  }
 }

src/main/java/com/elphel/imagej/tileprocessor/TwoQuadCLT.java

@@ -8315,7 +8315,6 @@ if (debugLevel > -100) return true; // temporarily !
 					}
 				}
 				
-				/*
 				boolean ok = quadCLT_main.extrinsicsCLT(
 						clt_parameters, // EyesisCorrectionParameters.CLTParameters           clt_parameters,
 						false, // adjust_poly,
@@ -8325,8 +8324,6 @@ if (debugLevel > -100) return true; // temporarily !
 // clear memory for main
 				quadCLT_main.tp.resetCLTPasses();
 				if (!ok) break;
-				*/
-
 			}
 			// Generate 4 main camera images and thumbnail
 			if (quadCLT_main.correctionsParameters.clt_batch_4img){