next versions, some cleanup

nn_ds_neibs16.py

 #!/usr/bin/env python3
-from numpy import float64
-from tensorflow.contrib.losses.python.metric_learning.metric_loss_ops import npairs_loss
-from debian.deb822 import PdiffIndex
+##from numpy import float64
+##from tensorflow.contrib.losses.python.metric_learning.metric_loss_ops import npairs_loss
+##from debian.deb822 import PdiffIndex
 
 __copyright__ = "Copyright 2018, Elphel, Inc."
 __license__   = "GPL-3.0+"
 __email__     = "andrey@elphel.com"
 
 
-from PIL import Image
+##from PIL import Image
 
 import os
 import sys
-import glob
+##import glob
 
 import numpy as np
-import itertools
+##import itertools
 
 import time
 
-import matplotlib.pyplot as plt
+##import matplotlib.pyplot as plt
 
 import shutil
 from threading import Thread
@@ -49,7 +49,7 @@ except IndexError:
     root_dir =  os.path.dirname(conf_file)
     
 print ("Configuration file: " + conf_file)
-parameters, dirs, files = qsf.parseXmlConfig(conf_file, root_dir)
+parameters, dirs, files, _ = qsf.parseXmlConfig(conf_file, root_dir)
 """
 Temporarily for backward compatibility
 """
@@ -221,7 +221,7 @@ if SPREAD_CONVERGENCE:
                                             
 else:
     outs, inp_weights =  qcstereo_network.networks_siam(
-                                            input=              corr2d_Nx325,
+                                            input_tensor=       corr2d_Nx325,
                                             input_global =      None,
                                             layout1 =           NN_LAYOUT1, 
                                             layout2 =           NN_LAYOUT2,
@@ -247,7 +247,7 @@ G_losses[0], _disp_slice, _d_gt_slice, _out_diff, _out_diff2, _w_norm, _out_wdif
               absolute_disparity =     ABSOLUTE_DISPARITY,
               use_confidence =         USE_CONFIDENCE, # True, 
               lambda_conf_avg =        0.01,
-              lambda_conf_pwr =        0.1,
+##              lambda_conf_pwr =        0.1,
               conf_pwr =               2.0,
               gt_conf_offset =         0.08,
               gt_conf_pwr =            2.0,
@@ -268,7 +268,7 @@ for n in range (1,len(partials)):
               absolute_disparity =     ABSOLUTE_DISPARITY,
               use_confidence =         USE_CONFIDENCE, # True, 
               lambda_conf_avg =        0.01,
-              lambda_conf_pwr =        0.1,
+#              lambda_conf_pwr =        0.1,
               conf_pwr =               2.0,
               gt_conf_offset =         0.08,
               gt_conf_pwr =            2.0,

qcstereo_functions.py

@@ -30,13 +30,14 @@ def print_time(txt="",end="\n"):
         txt +=" "
     print(("%s"+bcolors.BOLDWHITE+"at %.4fs (+%.4fs)"+bcolors.ENDC)%(txt,t-TIME_START,t-TIME_LAST), end = end, flush=True)
     TIME_LAST = t
+    
 def parseXmlConfig(conf_file, root_dir):
     tree = ET.parse(conf_file)
     root = tree.getroot()
     parameters = {}
     for p in root.find('parameters'):
         parameters[p.tag]=eval(p.text.strip())
-    globals    
+#    globals    
     dirs={}
     for p in root.find('directories'):
         dirs[p.tag]=eval(p.text.strip())
@@ -46,7 +47,11 @@ def parseXmlConfig(conf_file, root_dir):
     for p in root.find('files'):
         files[p.tag]=eval(p.text.strip())
 #    globals().update(parameters)
-    return parameters, dirs, files
+    dbg_parameters = {}
+    for p in root.find('dbg_parameters'):
+        dbg_parameters[p.tag]=eval(p.text.strip())
+
+    return parameters, dirs, files, dbg_parameters
 
 
 
@@ -84,7 +89,8 @@ def readTFRewcordsEpoch(train_filename):
     npy_dir_name = "npy"
     dirname = os.path.dirname(train_filename) 
     npy_dir = os.path.join(dirname, npy_dir_name)
-    filebasename, file_extension = os.path.splitext(train_filename)
+#    filebasename, file_extension = os.path.splitext(train_filename)
+    filebasename, _ = os.path.splitext(train_filename)
     filebasename = os.path.basename(filebasename)
     file_corr2d =           os.path.join(npy_dir,filebasename + '_corr2d.npy')
     file_target_disparity = os.path.join(npy_dir,filebasename + '_target_disparity.npy')
@@ -179,7 +185,7 @@ def add_neibs(npa_ext,radius):
     height = npa_ext.shape[0]-2*radius
     width =  npa_ext.shape[1]-2*radius
     side = 2 * radius + 1
-    size = side * side
+#    size = side * side
     npa_neib = np.empty((height,  width, side, side, npa_ext.shape[2]), dtype = npa_ext.dtype)
     for dy in range (side):
         for dx in range (side):
@@ -187,8 +193,8 @@ def add_neibs(npa_ext,radius):
     return npa_neib.reshape(height, width, -1)    
   
 def extend_img_to_clusters(datasets_img,radius, width): #  = 324):
-    side = 2 * radius + 1
-    size = side * side
+#    side = 2 * radius + 1
+#    size = side * side
     if len(datasets_img) ==0:
         return
     num_tiles = datasets_img[0]['corr2d'].shape[0]
@@ -210,7 +216,7 @@ def reformat_to_clusters(datasets_data, cluster_radius):
 
 def flip_horizontal(datasets_data, cluster_radius, tile_layers, tile_side):
     cluster_side = 2 * cluster_radius + 1
-    cluster_size = cluster_side * cluster_side
+#    cluster_size = cluster_side * cluster_side
     """
 TILE_LAYERS =        4
 TILE_SIDE =          9 # 7
@@ -238,8 +244,8 @@ TILE_SIZE =         TILE_SIDE* TILE_SIDE # == 81
         rec['target_disparity'] = target_disparity.reshape((target_disparity.shape[0],-1)) 
         rec['gt_ds'] =            gt_ds.reshape((gt_ds.shape[0],-1))
 
-def replace_nan(datasets_data, cluster_radius):
-    cluster_size = (2 * cluster_radius + 1) * (2 * cluster_radius + 1)
+def replace_nan(datasets_data): # , cluster_radius):
+#    cluster_size = (2 * cluster_radius + 1) * (2 * cluster_radius + 1)
 # Reformat input data
     for rec in datasets_data:
         if not rec is None:
@@ -259,7 +265,7 @@ def permute_to_swaps(perm):
         
 def shuffle_in_place(datasets_data, indx, period):
     swaps = permute_to_swaps(np.random.permutation(len(datasets_data)))
-    num_entries = datasets_data[0]['corr2d'].shape[0] // period
+#    num_entries = datasets_data[0]['corr2d'].shape[0] // period
     for swp in swaps:
         ds0 = datasets_data[swp[0]]
         ds1 = datasets_data[swp[1]]
@@ -279,9 +285,10 @@ def shuffle_chunks_in_place(datasets_data, tiles_groups_per_chunk):
     """
     Improve shuffling by preserving indices inside batches (0 <->0, ... 39 <->39 for 40 tile group batches)
     """
-    num_files = len(datasets_data)
+#    num_files = len(datasets_data)
     #chunks_per_file = datasets_data[0]['target_disparity']
-    for nf, ds in enumerate(datasets_data):
+#    for nf, ds in enumerate(datasets_data):
+    for ds in datasets_data:
         groups_per_file = ds['corr2d'].shape[0]
         chunks_per_file = groups_per_file//tiles_groups_per_chunk
         permut = np.random.permutation(chunks_per_file)
@@ -327,7 +334,8 @@ def zip_lvar_hvar(datasets_all_data, del_src = True):
                    'target_disparity': np.empty((recs[0]['target_disparity'].shape[0]*num_sets_to_combine,recs[0]['target_disparity'].shape[1]),dtype=np.float32),
                    'gt_ds':            np.empty((recs[0]['gt_ds'].shape[0]*num_sets_to_combine,           recs[0]['gt_ds'].shape[1]),dtype=np.float32)}
             
-            for nset, reci in enumerate(recs):
+#            for nset, reci in enumerate(recs):
+            for nset, _ in enumerate(recs):
                 rec['corr2d']          [nset::num_sets_to_combine] =  recs[nset]['corr2d'] 
                 rec['target_disparity'][nset::num_sets_to_combine] =  recs[nset]['target_disparity'] 
                 rec['gt_ds']           [nset::num_sets_to_combine] =  recs[nset]['gt_ds'] 
@@ -356,10 +364,10 @@ def initTrainTestData(
         max_files_per_group, # shuffling buffer for files
         two_trains,
         train_next):
-    datasets_train_lvar =  []
-    datasets_train_hvar =  []
-    datasets_train_lvar1 = []
-    datasets_train_hvar1 = []
+#    datasets_train_lvar =  []
+#    datasets_train_hvar =  []
+#    datasets_train_lvar1 = []
+#    datasets_train_hvar1 = []
     datasets_train_all = [[],[],[],[]]
     for n_train, f_train in enumerate(files['train']):
         if len(f_train) and ((n_train<2) or two_trains):
@@ -445,7 +453,8 @@ def readImageData(image_data,
              cluster_radius,
              width)
         if replace_nans:
-            replace_nan([image_data[indx]], cluster_radius)
+#            replace_nan([image_data[indx]], cluster_radius)
+            replace_nan([image_data[indx]])
             
     return image_data[indx]
                
@@ -477,7 +486,7 @@ def evaluateAllResults(result_files, absolute_disparity, cluster_radius):
     
 
 
-def result_npy_to_tiff(npy_path, absolute, fix_nan, insert_deltas=True):
+def result_npy_prepare(npy_path, absolute, fix_nan, insert_deltas=True):
     
     """
     @param npy_path full path to the npy file with 4-layer data (242,324,4) - nn_disparity(offset), target_disparity, gt disparity, gt strength
@@ -485,10 +494,9 @@ def result_npy_to_tiff(npy_path, absolute, fix_nan, insert_deltas=True):
     @param absolute - True - the first layer contains absolute disparity, False - difference from target_disparity
     @param fix_nan - replace nan in target_disparity with 0 to apply offset, target_disparity will still contain nan
     """
-    tiff_path = npy_path.replace('.npy','.tiff')
     data = np.load(npy_path) #(324,242,4) [nn_disp, target_disp,gt_disp, gt_conf]
     nn_out =            0
-    target_disparity =  1     
+#    target_disparity =  1     
     gt_disparity =      2     
     gt_strength =       3     
     if not absolute:
@@ -501,20 +509,28 @@ def result_npy_to_tiff(npy_path, absolute, fix_nan, insert_deltas=True):
         data = np.concatenate([data[...,0:4],data[...,0:2],data[...,0:2],data[...,4:]], axis = 2)
         data[...,6] -= data[...,gt_disparity] 
         data[...,7] -= data[...,gt_disparity]
-        for l in [4,5,6,7]:
+        for l in [2, 4, 5, 6, 7]:
             data[...,l] = np.select([data[...,gt_strength]==0.0, data[...,gt_strength]>0.0], [np.nan,data[...,l]])
         # All other layers - mast too
         for l in range(8,data.shape[2]):
             data[...,l] = np.select([data[...,gt_strength]==0.0, data[...,gt_strength]>0.0], [np.nan,data[...,l]])
+    return data        
+
+def result_npy_to_tiff(npy_path, absolute, fix_nan, insert_deltas=True):
     
-#        data[...,4] = np.select([data[...,3]==0.0, data[...,3]>0.0], [np.nan,data[...,4]])
-#       data[...,5] = np.select([data[...,3]==0.0, data[...,3]>0.0], [np.nan,data[...,5]])
+    """
+    @param npy_path full path to the npy file with 4-layer data (242,324,4) - nn_disparity(offset), target_disparity, gt disparity, gt strength
+           data will be written as 4-layer tiff, extension '.npy' replaced with '.tiff'
+    @param absolute - True - the first layer contains absolute disparity, False - difference from target_disparity
+    @param fix_nan - replace nan in target_disparity with 0 to apply offset, target_disparity will still contain nan
+    """
+    data = result_npy_prepare(npy_path, absolute, fix_nan, insert_deltas)
+    tiff_path = npy_path.replace('.npy','.tiff')
             
     data = data.transpose(2,0,1)
     print("Saving results to TIFF: "+tiff_path)
     imagej_tiffwriter.save(tiff_path,data[...,np.newaxis])        
 
-
 def eval_results(rslt_path, absolute,
                  min_disp =       -0.1, #minimal GT disparity
                  max_disp =       20.0, # maximal GT disparity

qcstereo_losses.py

@@ -4,7 +4,7 @@ __license__   = "GPL-3.0+"
 __email__     = "andrey@elphel.com"
 
 #from numpy import float64
-import numpy as np
+#import numpy as np
 import tensorflow as tf
 
 def smoothLoss(out_batch,                   # [batch_size,(1..2)] tf_result
@@ -76,7 +76,7 @@ def batchLoss(out_batch,                   # [batch_size,(1..2)] tf_result
               absolute_disparity =     False, #when false there should be no activation on disparity output ! 
               use_confidence =         False, 
               lambda_conf_avg =        0.01,
-              lambda_conf_pwr =        0.1,
+##              lambda_conf_pwr =        0.1,
               conf_pwr =               2.0,
               gt_conf_offset =         0.08,
               gt_conf_pwr =            1.0,
@@ -90,14 +90,14 @@ def batchLoss(out_batch,                   # [batch_size,(1..2)] tf_result
         Here confidence should be after relU. Disparity - may be also if absolute, but no activation if output is residual disparity
         """
         tf_lambda_conf_avg = tf.constant(lambda_conf_avg, dtype=tf.float32, name="tf_lambda_conf_avg")
-        tf_lambda_conf_pwr = tf.constant(lambda_conf_pwr, dtype=tf.float32, name="tf_lambda_conf_pwr")
-        tf_conf_pwr =        tf.constant(conf_pwr,        dtype=tf.float32, name="tf_conf_pwr")
+##        tf_lambda_conf_pwr = tf.constant(lambda_conf_pwr, dtype=tf.float32, name="tf_lambda_conf_pwr")
+##        tf_conf_pwr =        tf.constant(conf_pwr,        dtype=tf.float32, name="tf_conf_pwr")
         tf_gt_conf_offset =  tf.constant(gt_conf_offset,  dtype=tf.float32, name="tf_gt_conf_offset")
         tf_gt_conf_pwr =     tf.constant(gt_conf_pwr,     dtype=tf.float32, name="tf_gt_conf_pwr")
         tf_num_tiles =       tf.shape(gt_ds_batch)[0]
         tf_0f =              tf.constant(0.0,             dtype=tf.float32, name="tf_0f")
         tf_1f =              tf.constant(1.0,             dtype=tf.float32, name="tf_1f")
-        tf_maxw =            tf.constant(1.0,             dtype=tf.float32, name="tf_maxw")
+##        tf_maxw =            tf.constant(1.0,             dtype=tf.float32, name="tf_maxw")
         tf_disp_diff_cap2=   tf.constant(disp_diff_cap*disp_diff_cap,  dtype=tf.float32, name="disp_diff_cap2")
         tf_disp_diff_slope=  tf.constant(disp_diff_slope, dtype=tf.float32, name="disp_diff_slope")
         

qcstereo_network.py

@@ -4,7 +4,7 @@ __license__   = "GPL-3.0+"
 __email__     = "andrey@elphel.com"
 
 #from numpy import float64
-import numpy as np
+#import numpy as np
 import tensorflow as tf
 import tensorflow.contrib.slim as slim
 
@@ -13,15 +13,16 @@ def lrelu(x):
     return tf.maximum(x*0.2,x)
 #    return tf.nn.relu(x) 
 
-def sym_inputs8(inp):
+def sym_inputs8(inp, cluster_radius = 2):
     """
     get input vector [?:4*9*9+1] (last being target_disparity) and reorder for horizontal flip,
     vertical flip and transpose (8 variants, mode + 1 - hor, +2 - vert, +4 - transpose)
     return same lengh, reordered
     """
+    tile_side = 2 * cluster_radius + 1
     with tf.name_scope("sym_inputs8"):
         td =           inp[:,-1:] # tf.reshape(inp,[-1], name = "td")[-1]
-        inp_corr =     tf.reshape(inp[:,:-1],[-1,4,TILE_SIDE,TILE_SIDE], name = "inp_corr")
+        inp_corr =     tf.reshape(inp[:,:-1],[-1,4,tile_side,tile_side], name = "inp_corr")
         inp_corr_h =   tf.stack([-inp_corr  [:,0,:,-1::-1], inp_corr  [:,1,:,-1::-1], -inp_corr  [:,3,:,-1::-1], -inp_corr  [:,2,:,-1::-1]], axis=1, name = "inp_corr_h")
         inp_corr_v =   tf.stack([ inp_corr  [:,0,-1::-1,:],-inp_corr  [:,1,-1::-1,:],  inp_corr  [:,3,-1::-1,:],  inp_corr  [:,2,-1::-1,:]], axis=1, name = "inp_corr_v")
         inp_corr_hv =  tf.stack([ inp_corr_h[:,0,-1::-1,:],-inp_corr_h[:,1,-1::-1,:],  inp_corr_h[:,3,-1::-1,:],  inp_corr_h[:,2,-1::-1,:]], axis=1, name = "inp_corr_hv")
@@ -52,7 +53,7 @@ def sym_inputs8(inp):
                 tf.concat([tf.reshape(inp_corr_vt, [inp_corr.shape[0],-1]),td], axis=1,name = "out_corr_vt"),
                 tf.concat([tf.reshape(inp_corr_hvt,[inp_corr.shape[0],-1]),td], axis=1,name = "out_corr_hvt")]
         """
-        cl = 4 * TILE_SIDE * TILE_SIDE
+        cl = 4 * tile_side * tile_side
         return [tf.concat([tf.reshape(inp_corr,    [-1,cl]),td], axis=1,name = "out_corr"),
                 tf.concat([tf.reshape(inp_corr_h,  [-1,cl]),td], axis=1,name = "out_corr_h"),
                 tf.concat([tf.reshape(inp_corr_v,  [-1,cl]),td], axis=1,name = "out_corr_v"),
@@ -64,12 +65,13 @@ def sym_inputs8(inp):
 #                           inp_corr_h, inp_corr_v, inp_corr_hv, inp_corr_t, inp_corr_ht, inp_corr_vt, inp_corr_hvt]
     
 
-def network_sub(input,
+def network_sub(input_tensor,
                 input_global,  #add to all layers (but first) if not None
                 layout,
                 reuse,
-                sym8 = False):
-    last_indx = None;
+                sym8 = False,
+                cluster_radius = 2):
+#    last_indx = None;
     fc = []
     inp_weights = []
     for i, num_outs in enumerate (layout):
@@ -81,9 +83,9 @@ def network_sub(input,
                     inp = tf.concat([fc[-1], input_global], axis = 1)
                 fc.append(slim.fully_connected(inp,    num_outs, activation_fn=lrelu, scope='g_fc_sub'+str(i), reuse = reuse))
             else:
-               inp = input
+                inp = input_tensor
                 if sym8:
-                   inp8 = sym_inputs8(inp)
+                    inp8 = sym_inputs8(inp, cluster_radius)
                     num_non_sum = num_outs %  len(inp8) # if number of first layer outputs is not multiple of 8
                     num_sym8 =    num_outs // len(inp8) # number of symmetrical groups
                     fc_sym = []
@@ -111,12 +113,12 @@ def network_sub(input,
            
     return fc[-1], inp_weights
 
-def network_inter(input,
+def network_inter(input_tensor,
                   input_global,  #add to all layers (but first) if not None
                   layout,
                   reuse=False,
                   use_confidence=False):
-    last_indx = None;
+    #last_indx = None;
     fc = []
     for i, num_outs in enumerate (layout):
         if num_outs:
@@ -126,7 +128,7 @@ def network_inter(input,
                 else:
                     inp = tf.concat([fc[-1], input_global], axis = 1)
             else:
-               inp = input
+                inp = input_tensor
             fc.append(slim.fully_connected(inp,    num_outs, activation_fn=lrelu, scope='g_fc_inter'+str(i), reuse = reuse))
     if use_confidence:
         fc_out  = slim.fully_connected(fc[-1],     2, activation_fn=lrelu, scope='g_fc_inter_out', reuse = reuse)
@@ -135,7 +137,7 @@ def network_inter(input,
         #If using residual disparity, split last layer into 2 or remove activation and add rectifier to confidence only  
     return fc_out
 
-def networks_siam(input, # now [?,9,325]-> [?,25,325]
+def networks_siam(input_tensor, # now [?,9,325]-> [?,25,325]
                   input_global, # add to all layers (but first) if not None
                   layout1, 
                   layout2,
@@ -143,12 +145,13 @@ def networks_siam(input, # now [?,9,325]-> [?,25,325]
                   sym8 =        False,
                   only_tile =   None, # just for debugging - feed only data from the center sub-network
                   partials =    None,
-                  use_confidence=False):
+                  use_confidence=False,
+                  cluster_radius = 2):
                   
-    center_index = (input.shape[1] - 1) // 2 
+    center_index = (input_tensor.shape[1] - 1) // 2 
     with tf.name_scope("Siam_net"):
         inp_weights = []
-        num_legs =  input.shape[1] # == 25
+        num_legs =  input_tensor.shape[1] # == 25
         if partials is None:
             partials = [[True] * num_legs]
         inter_lists = [[] for _ in partials]
@@ -159,11 +162,12 @@ def networks_siam(input, # now [?,9,325]-> [?,25,325]
                     ig = None
                 else:
                     ig =input_global[:,i,:]
-                ns, ns_weights = network_sub(input[:,i,:],
+                ns, ns_weights = network_sub(input_tensor[:,i,:],
                                              ig, # input_global[:,i,:],
                                              layout= layout1,
                                              reuse= reuse,
-                                             sym8 = sym8)
+                                             sym8 = sym8,
+                                             cluster_radius = cluster_radius)
                 for n, partial in enumerate(partials):
                     if partial[i]:
                         inter_lists[n].append(ns)
@@ -178,7 +182,7 @@ def networks_siam(input, # now [?,9,325]-> [?,25,325]
             else:
                 ig =input_global[:,center_index,:]
             
-            outs.append(network_inter (input = tf.concat(inter_lists[n],
+            outs.append(network_inter (input_tensor = tf.concat(inter_lists[n],
                                                  axis=1,
                                                  name='inter_tensor'+str(n)),
                                        input_global =   [None, ig][inter_convergence], # optionally feed all convergence values (from each tile of a cluster)