Nantes Université

Skip to content
Extraits de code Groupes Projets
lg.py 80,6 ko
Newer Older
Richard Zanibbi's avatar
Richard Zanibbi a validé
################################################################
Richard Zanibbi's avatar
Richard Zanibbi a validé
#
# Authors: R. Zanibbi and H. Mouchere, 2012
# Copyright (c) 2012-2014 Richard Zanibbi and Harold Mouchere
Richard Zanibbi's avatar
Richard Zanibbi a validé
################################################################
import csv
import sys
import math
import copy
from io import StringIO
from collections import OrderedDict

from lgeval.src.smallGraph import SmallGraph
from lgeval.src.compareTools import cmpNodes, cmpEdges
Richard Zanibbi's avatar
Richard Zanibbi a validé

class Lg(object):
    """Class for bipartite graphs where the two node sets are identical, and
    multiple node and edge labels are permited. The graph and individual nodes
    and edges have associated values (e.g. weights/probabilities)."""

# Define graph data elements ('data members' for an object in the class)
    __slots__ = ('file','gweight','nlabels','elabels','error','absentNodes',\
                    'absentEdges','hiddenEdges', 'cmpNodes', 'cmpEdges','stringInput')

##################################
# Constructors (in __init__)
##################################
    def __init__(self,*args): 
        """Graph data is read from a CSV file or provided node and edge label
        dictionaries.  If invalid entries are found, the error flag is set to
        true, and graph input continues.  In .lg files, blank lines are
        ignored, and # may be used for comment lines in CSV graph files."""
        self.error = False
        self.gweight = 1.0
        # self.nlabels = {}
        # self.elabels = {}
        self.nlabels = OrderedDict()
        self.elabels = OrderedDict()
        self.absentNodes = set([])
        self.absentEdges = set([])
        self.hiddenEdges = {}
ayushkumarshah's avatar
ayushkumarshah a validé
        self.cmpNodes = cmpNodes
        self.cmpEdges = cmpEdges
        self.stringInput = False

        
        fileName = None
        nodeLabels = {}
        edgeLabels = {}
        
        validAsteriskEdges = set()
        invalidAsteriskNodes = set()

        if len(args) == 1:
            fileName = args[0]
            self.file = fileName # DEBUG: add filename for debugging purposes.
        elif len(args) == 2:
            nodeLabels = args[0]
            edgeLabels = args[1]

        if fileName == None:
            # CONSTRUCTOR 1: try to read in node and edge labels.
            self.file = None
            # Automatically convert identifiers and labels to strings if needed.
            for nid in list(nodeLabels):
                if not isinstance(nid, str):
                    nid = str(nid)

                newdict = {}
                for label in list(nodeLabels[nid]):
                    if not isinstance(nid, str):
                        label = str(label)
            
                    # Weights need to be floats.
                    if not isinstance( nodeLabels[nid][label], float ):
                        self.error = True
                        sys.stderr.write('  !! Invalid weight for node ' + nid + ', label \"' \
                                            + label +"\": " + str(nodeLabels[nid][label]) + "\n")
                    newdict[ label ] = nodeLabels[nid][label]
                self.nlabels[nid] = newdict

            # WARNING: self-edges are not detected if edge labels used
            # for initialization.
            for eid in list(edgeLabels):
                if not isinstance(eid[0], str) or not isinstance(eid[1],str):
                    eid[0] = str(eid[0])
                    eid[1] = str(eid[1])

                newdict = {}
                for label in list(edgeLabels[eid]):
                    if not isinstance(label, str):
                        label = str(label)
                    if not isinstance( edgeLabels[eid][label], float ):
                        self.error = True
                        sys.stderr.write('  !! Invalid weight for edge ' + str(eid) + ', label \"' \
                                            + label +"\": " + str(edgeLabels[eid][label]) + "\n")
                    newdict[ label ] = edgeLabels[eid][label]

                self.elabels[eid] = newdict
        else:
            # CONSTRUCTOR 2: Read graph data from CSV file.
            # RZ 2021: OR Read graph data from StringIO object in CSV format.
            MIN_NODE_ENTRY_LENGTH = 3
            MIN_EDGE_ENTRY_LENGTH = 4
            MIN_OBJECT_ENTRY_LENGTH = 5
            MIN_OBJECT_EDGE_ENTRY_LENGTH = 5
            try:
                    # If passed a StringIO object, read CSV from the string.
                    if type(fileName) == type( StringIO('') ):
                        fileReader = csv.reader( fileName )
                        self.stringInput = True
                    else:
                        fileReader = csv.reader(open(fileName))
            except:
                    # Create an empty graph if a file cannot be found.
                    # Set the error flag.
                    sys.stderr.write('  !! IO Error (cannot open): ' + fileName + '\n')
                    self.error = True
                    return
            objectDict = dict([])
            for row in fileReader:
                    # Skip blank lines.
                    if len(row) == 0 or len(row) == 1 and row[0].strip() == '':
                            continue

                    entryType = row[0].strip()
                    if entryType == 'N':
                            if len(row) < MIN_NODE_ENTRY_LENGTH:
                                    sys.stderr.write(' !! Invalid node entry length: ' +str(len(row))+\
                                                    '\n\t' + str(row) + '\n')
                                    self.error = True
                            else:
                                    nid = row[1].strip() # remove leading/trailing whitespace
                                    if nid in list(self.nlabels):
                                            nlabelDict = self.nlabels[ nid ]
                                            nlabel = row[2].strip()
                                            # if nlabel in nlabelDict:
                                                    # # Note possible error.
                                                    # sys.stderr.write(' !! Repeated node label entry ('\
                                                                    # + self.file + '): ' \
                                                                    # + '\n\t' + str(row) + '\n')
                                                    # self.error = True
                                            # Add (or replace) entry for the label.
                                            nlabelDict[ nlabel ] = float(row[3])
                                    else:
                                            # New primitive; create new dictionary for 
                                            # provided label (row[2]) and value (row[3])
                                            nid = row[1].strip()
                                            nlabel = row[2].strip()

                                            # Feb. 2013 - allow no weight to be provided.
                                            if len(row) > MIN_NODE_ENTRY_LENGTH:
                                                    self.nlabels[ nid ] = { nlabel : float(row[3]) }
                                            else:
                                                    self.nlabels[ nid ] = { nlabel : 1.0 }

                    elif entryType == 'E':
                            if len(row) < MIN_EDGE_ENTRY_LENGTH:
                                    sys.stderr.write(' !! Invalid edge entry length: ' +str(len(row))+\
                                                    '\n\t' + str(row) + '\n')
                                    self.error = True
                            else:
                                    primPair = ( row[1].strip(), row[2].strip() )
                                    #self to self edge = error
                                    if primPair[0] == primPair[1]:
                                            sys.stderr.write('  !! Invalid self-edge (' +
                                                            self.file + '):\n\t' + str(row) + '\n')
                                            self.error = True
                                            nid = primPair[0]
                                            if nid in list(self.nlabels):
                                                    nlabelDict = self.nlabels[ nid ]
                                                    nlabel = row[3].strip()
                                                    # if nlabel in nlabelDict:
                                                            # # Note possible error.
                                                            # sys.stderr.write(' !! Repeated node label entry ('\
                                                                    # + self.file + '): ' \
                                                                    # + '\n\t' + str(row) + '\n')
                                            # Add (or replace) entry for the label.
                                            nlabelDict[ nlabel ] = float(row[4])

                                    #an edge already existing, add a new label
                                    elif primPair in list(self.elabels):
                                            elabelDict = self.elabels[ primPair ]
                                            elabel = row[3].strip()
                                            # if elabel in elabelDict:
                                                    # # Note possible error.
                                                    # sys.stderr.write(' !! Repeated edge label entry (' \
                                                                    # + self.file + '):\n\t' + str(row) + '\n')
                                                    # self.error = True
                                            if elabel == '*':
                                                    # if using old fashion segmentation label, convert it by finding the (only) node label
                                                    if primPair[0] in self.nlabels and primPair[1] in self.nlabels and \
                                                    self.nlabels[ primPair[0]] == self.nlabels[ primPair[1]]:
                                                            elabel =  list(self.nlabels[ primPair[0]])[0]
                                                            
                                                            validAsteriskEdges.add( primPair )

                                                    else:
                                                            sys.stderr.write(' !! * edge used with ambiguous node labels (' \
                                                                    + str(self.nlabels[ primPair[0]]) + ' vs. ' \
                                                                    + str(self.nlabels[ primtPair[1]]) + ') in ' \
                                                                    + self.file + '):\n\t' + ", ".join(row) + '\n')
                                                            
                                                            # RZ: Oct. 14 - cheap and dirty correction.
                                                            elabel = 'MergeError'
                                                            self.nlabels[ primPair[0] ] = { elabel : 1.0 }
                                                            self.nlabels[ primPair[1] ] = { elabel : 1.0 }
                                                            self.error = True

                                                            invalidAsteriskNodes.add( primPair[0] )
                                                            invalidAsteriskNodes.add( primPair[1] )
            
                                            # Add (or replace) entry for the label.
                                            # Feb. 2013 - allow no weight.
                                            if len(row) > MIN_EDGE_ENTRY_LENGTH:
                                                    elabelDict[ elabel ] = float(row[4])
                                            else:
                                                    elabelDict[ elabel ] = 1.0
                                    else:
                                            # Add new edge label entry for the new edge label
                                            # as a dictionary.
                                            primPair = ( row[1].strip(), row[2].strip() )
                                            elabel = row[3].strip()
                                            if elabel == '*':
                                                    # if using old fashion segmentation label, convert it by finding the (only) node label
                                                    if primPair[0] in self.nlabels and primPair[1] in self.nlabels and \
                                                    self.nlabels[ primPair[0]] == self.nlabels[ primPair[1]]:
                                                            elabel = list(self.nlabels[ primPair[0]])[0]
                                                            validAsteriskEdges.add( primPair )

                                                    else:
                                                            sys.stderr.write(' !! * edge used with ambiguous node labels (' \
                                                                    + str(self.nlabels[ primPair[0]]) + ' vs. ' \
                                                                    + str(self.nlabels[ primPair[1]]) + ') in ' \
                                                                    + self.file + '):\n\t' + ", ".join(row) + '\n')
                                                            
                                                            elabel = 'MergeError'
                                                            self.nlabels[ primPair[0] ] = { elabel : 1.0 }
                                                            self.nlabels[ primPair[1] ] = { elabel : 1.0 }
                                                            self.error = True

                                                            invalidAsteriskNodes.add( primPair[0] )
                                                            invalidAsteriskNodes.add( primPair[1] )

                                            self.elabels[ primPair ] = { elabel : float(row[4]) }
                    elif entryType == 'O':
                            if len(row) < MIN_OBJECT_ENTRY_LENGTH:
                                    sys.stderr.write(' !! Invalid object entry length: '+str(len(row))+\
                                                    '\n\t' + str(row) + '\n')
                                    self.error = True
                            else:
                                    rawnodeList = row[4:] # get all other item as node id
                                    oid =  row[1].strip()
                                    nlabel =  row[2].strip()
                                    nValue =  float(row[3].strip())
                                    nodeList = []
                                # add all nodes
                                    for n in rawnodeList:
                                            nid = n.strip()
                                            nodeList.append(nid)
                                            if nid in list(self.nlabels):
                                                    nlabelDict = self.nlabels[ nid ]
                                                    
                                                    # Add (or replace) entry for the label.
                                                    nlabelDict[ nlabel ] = nValue
                                            else:
                                                    # New primitive; create new dictionary for 
                                                    # provided label and value 	
                                                    # Feb. 2013 - allow no weight to be provided.
                                                    self.nlabels[ nid ] = { nlabel : nValue }
                                    #save the nodes of this object
                                    objectDict[oid] = nodeList
                                    #add all edges
                                    for nid1 in nodeList:
                                            #nid1 = n1.strip()
                                            for nid2 in nodeList:
                                                    #nid2 = n2.strip()
                                                    if nid1 != nid2:
                                                            primPair = ( nid1, nid2 )
                                                            elabel = nlabel 
                                                            if primPair in list(self.elabels):
                                                                    elabelDict = self.elabels[ primPair ]
                                                                    
                                                                    # Add (or replace) entry for the label.
                                                                    elabelDict[ elabel ] = nValue
                                                            else:
                                                                    # Add new edge label entry for the new edge label
                                                                    # as a dictionary.
                                                                    self.elabels[ primPair ] = { elabel : nValue }

                    elif entryType == 'R' or entryType == 'EO':
                            if len(row) < MIN_OBJECT_EDGE_ENTRY_LENGTH:
                                    sys.stderr.write(' !! Invalid object entry length: ' +str(len(row))+\
                                                    '\n\t' + str(row) + '\n')
                                    self.error = True
                            else:
                                    oid1 = row[1].strip()
                                    oid2 = row[2].strip()
                                    elabel = row[3].strip()
                                    eValue = float(row[4].strip())
                                    validRelationship = True

                                    if not oid1 in objectDict:
                                        sys.stderr.write(' !! Invalid object id: "' + oid1+\
                                                        '" - IGNORING relationship:\n\t' + str(row) + '\n')
                                        self.error = True
                                        validRelationship = False
                                    if not oid2 in objectDict:
                                        sys.stderr.write(' !! Invalid object id: "' + oid2+\
                                                        '" - IGNORING relationship:\n\t' + str(row) + '\n')
                                        self.error = True
                                        validRelationship = False
                                    if validRelationship:
                                        nodeList1 = objectDict[oid1] # get all other item as node id
                                        nodeList2 = objectDict[oid2] # get all other item as node id

                                        for nid1 in nodeList1:
                                                for nid2 in nodeList2:
                                                        if nid1 != nid2:
                                                                primPair = ( nid1, nid2 )
                                                                if primPair in list(self.elabels):
                                                                        elabelDict = self.elabels[ primPair ]
                                                                        
                                                                        # Add (or replace) entry for the label.
                                                                        elabelDict[ elabel ] = eValue
                                                                else:
                                                                        # Add new edge label entry for the new edge label
                                                                        # as dictionary.
                                                                        self.elabels[ primPair ] = { elabel : eValue }
                                                        else:			
                                                                sys.stderr.write('  !! Invalid self-edge (' +
                                                                self.file + '):\n\t' + str(row) + '\n')
                                                                self.error = True

                    # DEBUG: complaints about empty lines here...
                    elif len(entryType.strip()) > 0 and entryType.strip()[0] == '#':
                        # Ignore lines with comments.
                        pass
                    else:
                        sys.stderr.write('  !! Invalid graph entry type (expected N, E, O, R or EO): ' \
                                        + str(row) + '\n')
                        self.error = True

        # Add any implicit nodes in edges explicitly to the hash table
        # containing nodes. The 'nolabel' label is '_'.
        anonNode = False
        anodeList = []
        for elabel in list(self.elabels):
            nid1 = elabel[0]
            nid2 = elabel[1]

            if not nid1 in list(self.nlabels):
                    self.nlabels[ nid1 ] = { '_' : 1.0 }
                    anodeList = anodeList + [ nid1 ]
                    anonNode = True
            if not nid2 in list(self.nlabels):
                    self.nlabels[ nid2 ] = { '_' : 1.0 }
                    anodeList = anodeList + [ nid2 ]
                    anonNode = True
        if anonNode:
            sys.stderr.write('  ** Anonymous labels created for:\n\t' \
                    + str(anodeList) + '\n')


        # RZ Oct. 2014: add invalid merge edges and node labels where missing.
        #    This catches when a valid * edge is connected to an invalid one,
        #    relabeling the edge.
        invalidAsteriskNodeList = sorted( list(invalidAsteriskNodes) )
        while len(invalidAsteriskNodeList) > 0:
            # Remove last element from the list.
            nextPrimId = invalidAsteriskNodeList.pop()
            
            # Linear traversal for matches (a 'region growing' algorithm)
            # Add a traversal each time a new connected edge is found.
            # NOTE: this will not add edges missing in the input (e.g.
            #  if '*' is defined in one direction but not the other.
            for (parent, child) in validAsteriskEdges:
                otherId = None
                if parent == nextPrimId:
                    otherId = child
                if child == nextPrimId:
                    otherId = parent

                if otherId != None:
                    if not otherId in invalidAsteriskNodes:
                        invalidAsteriskNodes.add( otherId )
                        invalidAsteriskNodeList.append( otherId )

                    self.nlabels[ otherId ] = { 'MergeError' : 1.0 }
                    self.elabels[ (parent, child) ] = { 'MergeError' : 1.0 }

##################################
# String, CSV output
##################################
    def __str__(self):
            nlabelcount = 0
            elabelcount = 0
            for nid in list(self.nlabels):
                    nlabelcount = nlabelcount + len(list(self.nlabels[nid]))
            for eid in list(self.elabels):
                    elabelcount = elabelcount + len(list(self.elabels[eid]))

            return 'Nodes: ' + str(len(list(self.nlabels))) \
                            + ' (labels: ' + str(nlabelcount) \
                            + ')   Edges: ' + str(len(list(self.elabels))) \
                            + ' (labels: ' + str(elabelcount) \
                            + ')   Error: ' + str(self.error)


    def csvObject(self):
            """Construct CSV data file using object-relationship format. Currently 
            weight values are only placeholders (i.e. 1.0 is always used)."""
            outputString = ""

            (segmentPrimitiveMap, primitiveSegmentMap, rootSegments, \
                            segmentEdges) = self.segmentGraph()

            # Write the file name if file is used.
            if self.stringInput == False:
                outputString += "# " + os.path.split(self.file)[1]
                outputString += "\n\n"

            # Write number of objects and format information.
            # Output object information.
            outputString += "# " + str(len(list(segmentPrimitiveMap))) + " Objects"
            outputString += "\n"
            outputString += "# FORMAT: O, Object ID, Label, Weight, [ Primitive ID List ]"
            outputString += "\n"

            for objectId in sorted( list(segmentPrimitiveMap) ):
                    for label in sorted(segmentPrimitiveMap[objectId][1]):
                            outputString += "O, " + objectId + ", " + label + ", 1.0"
                            for primitiveId in sorted( segmentPrimitiveMap[ objectId ][ 0 ] ):
                                    outputString += ", " + primitiveId 
                            outputString += "\n"

            # Write number of relationships and format information.
            # Write relationship information.
            outputString += "\n"
            outputString += "# " + str( len(list(segmentEdges)) ) + " Relationships (Pairs of Objects)"
            outputString += "\n"
            outputString += "# FORMAT: R, Object ID (parent), Object ID (child), Label, Weight" 
            outputString += "\n"

            for (parentObj, childObj) in sorted( list(segmentEdges) ):
                    for relationship in sorted( list(segmentEdges[ (parentObj, childObj) ]) ):
                            outputString += "R, " + parentObj + ", " + childObj + ", " 
                            outputString += relationship + ", 1.0"
                            outputString += "\n"

            return outputString


    def csv(self, sort=True):
            """Construct CSV data file representation as a string."""
            # NOTE: currently the graph value is not being stored...
            sstring = ''
            nlist = []
            elist = []
            for nkey in list(self.nlabels):
                    nodeLabels = self.nlabels[nkey]
                    for nlabel in list(nodeLabels):
                            nstring = 'N,' + nkey + ',' + nlabel + ',' + \
                                            str(nodeLabels[nlabel]) + '\n'
                            nlist = nlist + [ nstring ]

            for npair in list(self.elabels):
                    edgeLabels = self.elabels[npair]
                    for elabel in list(edgeLabels):
                            estring = 'E,' + npair[0] + ',' + npair[1] + ',' + elabel + ',' + \
                                            str(edgeLabels[ elabel ]) + '\n'
                            elist = elist + [ estring ]

            if sort:
                # Sort the node and edge strings lexicographically.
                # NOTE: this means that '10' precedes '2' in the sorted ordering
                nlist.sort()
                elist.sort() 

            if self.stringInput == False:
                sstring += '# ' + os.path.split(self.file)[1] + '\n\n' 

485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000
            sstring += '# ' + str(len(nlist)) + ' Nodes\n'
            sstring += "# FORMAT: N, Primitive ID, Label, Weight\n"
            for nstring in nlist:
                    sstring = sstring + nstring
            sstring += "\n"

            sstring += '# ' + str(len(elist)) + ' Edges\n'
            sstring += '# FORMAT: E, Primitive ID (parent), Primitive ID (child), Label, Weight\n'
            for estring in elist:
                    sstring = sstring + estring
            
            return sstring

##################################
# Construct segment-based graph
# for current graph state
##################################
    def segmentGraph(self):
            """Return dictionaries from segments to strokes, strokes to segments,
            segments without parents, and edges labeled as segment (w. symbol label)."""
            primitiveSegmentMap = {}
            segmentPrimitiveMap = {}
            #noparentSegments = []
            segmentEdges = {}  # Edges between detected objects (segments)

            self.hideUnlabeledEdges()

            # Note: a segmentation edge in either direction merges a primitive pair.
            primSets = {}
            for node,labs in self.nlabels.items():
                    primSets[node] = {}
                    for l in labs:
                            (cost,_)=self.cmpNodes([l],[])
                            if(cost > 0):
                                    primSets[node][l] = set([node])
                    #if len(primSets[node]) == 0:
                    #	primSets[node]['_'] = set([node]) #at least one empty label
            for (n1, n2) in list(self.elabels):
                    commonLabels = set(list(self.nlabels[n1])).intersection(list(self.nlabels[n2]),list(self.elabels[(n1,n2)]))
                    for l in commonLabels:
                            #check if this label is interesting or not => compare to 'nothing', if there is not error, it means it is not interesting
                            (cost,_)=self.cmpNodes([l],[])
                            if(cost > 0):
                                    primSets[n1][l].add(n2)
                                    primSets[n2][l].add(n1)

            # NOTE: Segments can have multiple labels
            # A primitive can belong to several different
            # segments with different sets of primitives with different labels.
            # but there is only one segment with the same label attached to each primitive.
            i = 0
            segmentList = []
            rootSegments = set([])
            
            # For each label associated with each primitive, there is a possible object/segment
            for primitive,segments in primSets.items():
                    if not primitive in primitiveSegmentMap:
                            primitiveSegmentMap[ primitive ] = {}
                    for lab in list(segments):
                            alreadySegmented = False
                            for j in range(len(segmentList)):
                                    if segments[lab] == segmentList[j]["prim"]:
                                            if not primitive in primitiveSegmentMap:
                                                    primitiveSegmentMap[ primitive ] = {}
                                            primitiveSegmentMap[ primitive ][lab] = 'Obj' + str(j)
                                            alreadySegmented = True
                                            if lab not in segmentList[j]["label"]:
                                                    segmentPrimitiveMap[  'Obj' + str(j) ][1].append(lab)
                                                    segmentList[j]["label"].add(lab)
                                            break

                            if not alreadySegmented:
                                    # Add the new segment.
                                    newSegment = 'Obj' + str(i)
                                    segmentList = segmentList + [ {"label":{lab},"prim":primSets[primitive][lab]} ]
                                    segmentPrimitiveMap[ newSegment ] = (segments[lab],[lab])
                                    primitiveSegmentMap[ primitive ][lab] = newSegment
                                    rootSegments.add(newSegment)
                                    i += 1

            # Identify 'root' objects/segments (i.e. with no incoming edges),
            # and edges between objects. **We skip segmentation edges.
            for (n1, n2), elabs in self.elabels.items():
                    segment1 = primitiveSegmentMap[n1]
                    segment2 = primitiveSegmentMap[n2]
                    
                    #for all possible pair of segments with these two primitives, look for the effective relation labels
                    possibleRelationLabels = set(list(elabs)).difference(list(self.nlabels[n1]),list(self.nlabels[n2]))
                    if len(possibleRelationLabels) != 0:
                            #for all pair of labels
                            for l1,pset1 in segment1.items():
                                    for l2, pset2 in segment2.items():
                                            #if not in the same seg
                                            if pset1 != pset2:
                                                    #look for the label which is common for all primitive pair in the two segments
                                                    theRelationLab = possibleRelationLabels
                                                    for p1 in primSets[n1][l1]:
                                                            for p2 in primSets[n2][l2]:
                                                                    if(p1,p2) in self.elabels:
                                                                            theRelationLab &= set(list(self.elabels[(p1,p2)]))
                                                                    else:
                                                                            theRelationLab = set([]) # it should be a clique !
                                                                    if len(theRelationLab) == 0:
                                                                            break
                                                            if len(theRelationLab) == 0:
                                                                    break
                                                    # there is a common relation if theRelationLab is not empty
                                                    if len(theRelationLab) != 0:
                                                            #we can remove seg2 from the roots
                                                            if pset2 in rootSegments:
                                                                    rootSegments.remove(pset2)
                                                            #print (str((n1, n2))+ " => " + str(( pset1,  pset2)) + "  = " + str(theRelationLab))
                                                            for label in theRelationLab:
                                                                    #check if this label is interesting or not => compare to 'nothing', if there is not error, it means it is not interesting
                                                                    (cost,_)=self.cmpNodes([label],[])
                                                                    if(cost > 0):
                                                                            if ( pset1,  pset2) in segmentEdges:
                                                                                    if label in segmentEdges[ ( pset1,  pset2) ]:
                                                                                            # Sum weights for repeated labels
                                                                                            segmentEdges[ ( pset1,  pset2)][label] += \
                                                                                                            self.elabels[(n1,n2)][label]
                                                                                    else:
                                                                                            # Add unaltered weights for new edge labels
                                                                                            segmentEdges[ ( pset1,  pset2) ][label] = \
                                                                                                            self.elabels[(n1,n2)][label]
                                                                            else:
                                                                                    segmentEdges[ ( pset1, pset2) ] = {}
                                                                                    segmentEdges[ ( pset1, pset2) ][label] = \
                                                                                                    self.elabels[(n1,n2)][label]

            self.restoreUnlabeledEdges()

            return (segmentPrimitiveMap, primitiveSegmentMap, list(rootSegments), \
                            segmentEdges)


##################################
# Metrics and Graph Differences
##################################
    def compareSegments(self, lg2):
            """Compute the number of differing segments, and record disagreements.
            The primitives in each graph should be of the same number and names
            (identifiers). Nodes are merged that have identical (label,value)
            pairs on nodes and all incoming and outgoing edges."""
            (sp1, ps1, _, sre1) = self.segmentGraph()
            (sp2, ps2, _, sre2) = lg2.segmentGraph()

            allNodes = set(list(ps1))
            assert allNodes == set(list(ps2))

            edgeDiffCount = 0
            edgeDiffClassCount = 0
            segDiffs = {}
            correctSegments = set([])
            correctSegmentsAndClass = set([])
            undirDiffClassSet = set([])
            
            # List and count errors due to segmentation.
            # Use cmpNodes to compare the labels of symbols.
            # Idea : build the sub graph with the current primitive as center and only 
            for primitive in list(ps1):
                    edgeFromP1 = {}
                    edgeFromP2 = {}
                    for (lab1,seg1) in ps1[primitive].items():
                            for p in sp1[seg1][0]:
                                    # DEBUG (RZ): this is producing a primitive edge-level count:
                                    # do not count segment edges that are undefined (e.g. in one direction,
                                    # but not the other)
                                    if p != primitive and (p,primitive) in list(self.elabels) and \
                                                    lab1 in list(self.elabels[ (p,primitive) ]):
                                            if p in edgeFromP1:
                                                    edgeFromP1[p].append(lab1)
                                            else:  
                                                    edgeFromP1[p] = [lab1]

                    for (lab2,seg2) in ps2[primitive].items():
                            for p in sp2[seg2][0]:
                                    # DEBUG (RZ) - see DEBUG comment above.
                                    if p != primitive and (p,primitive) in list(lg2.elabels) and \
                                                    lab2 in list(lg2.elabels[ (p, primitive) ]):
                                            if p in edgeFromP2:
                                                    edgeFromP2[p].append(lab2)
                                            else:
                                                    edgeFromP2[p] = [lab2]

                    # Compute differences in edge labels with cmpNodes (as they are symbol labels)
                    diff1 = set([])
                    diff2 = set([])
                    
                    # first add differences for shared primitives
                    commonPrim = set(list(edgeFromP1)).intersection(list(edgeFromP2))
                    for p in commonPrim:
                            (cost,diff) = self.cmpNodes(edgeFromP1[p], edgeFromP2[p])
                            edgeDiffCount = edgeDiffCount + cost

                            # RZ June 2015: Record edges that are specifically valid merges with disagreeing labels.
                            #     Also record sets of undirected edges that disagree.
                            for (l1,l2) in diff:
                                    if l1 in list(self.nlabels[p]) and l2 in list(lg2.nlabels[p]):
                                            edgeDiffClassCount += 1
                                    
                                    # RZ: we do not have a *segmentation* difference if corresponding segm.
                                    #     edges have a label.
                                    elif cost > 0:
                                            diff1.add(p)
                                            diff2.add(p)

                                    if not (p, primitive) in undirDiffClassSet and not (primitive, p) in undirDiffClassSet:
                                            undirDiffClassSet.add( (primitive, p) )

                    #then add differences for primitives which are not in the other set
                    for p in (set(list(edgeFromP1)) - commonPrim):
                            (cost,diff) = self.cmpNodes(edgeFromP1[p], [])
                            edgeDiffCount = edgeDiffCount + cost
                            diff1.add(p)
                                    
                    for p in (set(list(edgeFromP2)) - commonPrim):
                            (cost,diff) = self.cmpNodes(edgeFromP2[p], [])
                            edgeDiffCount = edgeDiffCount + cost
                            diff2.add(p)
                                    

                    # Only create an entry where there are disagreements.
                    if len(diff1) + len(diff2) > 0:
                            segDiffs[primitive] = ( diff1, diff2 )
                    
            # RZ: Oct. 2014 - replacing method used to evaluate segmentation. Also
            #     add checks for segments in the target being disjoint.
            #
            # Objects are defined by a set of primitives, plus a label. 
            # NOTE: This currently will support mutliple labels, but will lead to invalid
            #   "Class/Det" values in 00_Summary.txt if there are multiple labels.
            targets = {}

            # RZ: Add mapping from primitive list to object ids for direct lookup.
            targetObjIds = {}
            matchedTargets = set()
            for ObjID in list(sp2):
                    # Skip absent nodes - they are not valid targets.
                    if 'ABSENT' not in sp2[ ObjID ][ 1 ]:
                            # Convert primitive set to a sorted tuple list.
                            primitiveTupleList = tuple( sorted( list( sp2[ ObjID ][ 0 ] ) ) )
                    
                            # Store target label in targets dict, matches in matchedTargets dict (false init.)
                            targets[ primitiveTupleList ] = sp2[ ObjID][1]
                            targetObjIds[ primitiveTupleList ] = ObjID
            
            # Look for matches.
            # Do *not* allow a primitive set to be matched more than once.
            for ObjID in list(sp1):
                    # HACK (RZ): DEBUG - was not checking whether matched objects were
                    #               missing before absent nodes were added.
                    if 'ABSENT' in sp1[ ObjID ][ 1 ]:
                            continue

                    primitiveTupleList = tuple( sorted( list(sp1[ObjID][ 0 ] )))
                    if primitiveTupleList in list(targets) \
                                    and not primitiveTupleList in matchedTargets:
                            matchedTargets.add( primitiveTupleList )
                            correctSegments.add( ObjID )
                            
                            # Obtain matching labels. Create list of correct (segmentId, label) pairs
                            # for *all* matching labels.
                            # DEBUG: empty lists were being matched! Added test for empty matches.
                            # WARNING: Only guaranteed to work for single labels.
                            outputLabels = set(sp1[ ObjID ][ 1 ])
                            matchingLabels = list( outputLabels.intersection( targets[ primitiveTupleList ] ) )
                            if len(matchingLabels) > 0:
                                    ObjIDRepeats = [ObjID] * len(matchingLabels)
                                    correctSegmentsAndClass.add( tuple( zip(ObjIDRepeats, list(matchingLabels))))

            # Compute total number of object classifications (recognition targets)
            nbSegmClass = 0
            for (_,labs) in sp2.items():
                    nbSegmClass += len(labs[1])

            # Compute the specific 'object-level' graph edges that disagree, at the
            # level of primitive-pairs. 
            segRelErrors = 0
            correctSegRels = 0
            correctSegRelLocations = 0
            primRelEdgeDiffs = {}
            
            # Iterate over object relationships in the output graph.
            for thisPair in list(sre1):
                    misLabeled = False
                    falsePositive = False

                    thisParentIds = set(sp1[ thisPair[0] ][0])
                    thisChildIds = set(sp1[thisPair[1] ][0])
                    
                    # RZ (June 2015): Obtain names for correct segments in target graph (lg2)
                    primitiveTupleListParent = tuple( sorted( list( thisParentIds )))
                    primitiveTupleListChild =  tuple( sorted( list ( thisChildIds )))
                    targetObjNameParent = None
                    targetObjNameChild = None

                    if primitiveTupleListParent in list(targetObjIds):
                            targetObjNameParent = targetObjIds[ primitiveTupleListParent ]
                    if primitiveTupleListChild in list(targetObjIds):
                            targetObjNameChild = targetObjIds[ primitiveTupleListChild ]
                    
                    # Check whether the objects are correctly segmented by their object identifiers
                    if not ( thisPair[0] in correctSegments and  thisPair[1] in correctSegments):
                            # Avoid counting mis-segmented objects as having valid relationships
                            falsePositive = True
                    elif not ( targetObjNameParent, targetObjNameChild ) in list(sre2):
                            # Check that there is an edge between these objects in the target graph.
                            falsePositive = True
                    else:
                            # RZ (June, 2015): Compare labels directly on relation edges.
                            # WARNING: This checks that *all* labels are identical. Fine for single labels.
                            if not sorted( list(sre1[ thisPair ]) ) == \
                                            sorted( list(sre2[ ( targetObjNameParent, targetObjNameChild )]) ):
                                    misLabeled = True
                                    
                    # NOTE: assumes single labels on primitives.
                    # primRelEdgeDiffs records which object pairs have incorrect labels.
                    if falsePositive or misLabeled:
                            self.error = True
                            segRelErrors += 1
                            primRelEdgeDiffs[ thisPair ] = [ ('Error',1.0) ]
                    else:
                            correctSegRels += 1
                    
                    # Count correct relationship structures/locations.
                    if not falsePositive:
                            correctSegRelLocations += 1

            # Compute object counts *without* inserted absent nodes.
            lg2.removeAbsent()
            self.removeAbsent()

            (sp2orig, ps2orig, _, sre2orig) = lg2.segmentGraph()
            (sp1orig, ps1orig, _, sre1orig) = self.segmentGraph()
            
            nLg2Objs = len(list(sp2orig)) 
            nLg1Objs = len(list(sp1orig)) 

            # For input file, need to compare against all objects after including
            # missing/additional absent nodes and edges.
            nLg1ObjsWithAbsent = len(list(sp1))

            lg2.addAbsent(self)
            self.addAbsent(lg2)
            
            # RZ (Oct. 2014) Adding indicator variables for different correctness scenarios.
            hasCorrectSegments = 1 if len(correctSegments) == nLg2Objs and \
                            len(correctSegments) == nLg1ObjsWithAbsent else 0
            hasCorrectSegmentsAndLabels = 1 if len(correctSegmentsAndClass) == nLg2Objs and \
                            len(correctSegmentsAndClass) == nLg1ObjsWithAbsent else 0
            
            hasCorrectRelationLocations = 1 if correctSegRelLocations == len(list(sre1)) and \
                            correctSegRelLocations == len(list(sre2)) else 0
            hasCorrectRelationsAndLabels =  1 if correctSegRels == len(list(sre1)) and \
                            correctSegRels == len(list(sre2)) else 0
            
            hasCorrectStructure = hasCorrectRelationLocations and hasCorrectSegments
            
            # Compile vector of (name, value) metric pairs.
            metrics = [
                    ("edgeDiffClassCount", edgeDiffClassCount),
                    ("undirDiffClassCount", len(undirDiffClassSet)),
                    
                    ("nSeg", nLg2Objs),
                    ("detectedSeg", nLg1Objs),
                    ("dSegRelEdges", len(list(sre1))),
                    ("CorrectSegments", len(correctSegments)),
                ("CorrectSegmentsAndClass", len(correctSegmentsAndClass)),
                    ("ClassError", nbSegmClass - len(correctSegmentsAndClass)), 
                    ("CorrectSegRels",correctSegRels),
                    ("CorrectSegRelLocations",correctSegRelLocations),
                    ("SegRelErrors", segRelErrors),
                    
                    ("hasCorrectSegments", hasCorrectSegments),
                    ("hasCorrectSegLab", hasCorrectSegmentsAndLabels), 
                    ("hasCorrectRelationLocations", hasCorrectRelationLocations),
                    ("hasCorrectRelLab", hasCorrectRelationsAndLabels),
                    ("hasCorrectStructure", hasCorrectStructure) ]

            # RZ: June 2015 - need to subtract misclassified edges from non-matching edges
            # to obtain correct "Delta S" (D_S) Hamming distance for mismatched
            # segmentation edges.
            segEdgeMismatch = edgeDiffCount - edgeDiffClassCount

            return (segEdgeMismatch, segDiffs, correctSegments, metrics, primRelEdgeDiffs)

    def compare(self, lg2):
            """Returns: 1. a list of (metric,value) pairs,
            2. a list of (n1,n2) node disagreements, 3. (e1,e2) pairs
            for edge disagreements, 4. dictionary from primitives to
            disagreeing segment graph edges for (self, lg2). Node and 
            edge labels are compared using label sets without values, and
            *not* labels sorted by value."""
            metrics  = []
            nodeconflicts = []
            edgeconflicts = []

            # HM: use the union of all node labels instead of only lg2 ones
            #     it changes the nlabelMismatch, nodeClassError and so D_C and all rates values
            allNodes = set(list(lg2.nlabels)).union(list(self.nlabels))
            numNodes = len(allNodes)
            (sp2, ps2, _, sre2) = lg2.segmentGraph()
            nSegRelEdges = len(sre2)

            # Handle case of empty graphs, and missing primitives.
            # SIDE EFFECT: 'ABSENT' nodes added to each graph.
            self.matchAbsent(lg2)

            # METRICS
            # Node and edge labels are considered as sets.
            nlabelMismatch = 0
            numEdges = numNodes * (numNodes - 1)  # No self-edges.
            numLabels = numNodes + numEdges
            elabelMismatch = 0

            # Mismatched nodes.
            nodeClassError = set()
            for nid in allNodes: #list(self.nlabels):
                    (cost,errL) = self.cmpNodes(list(self.nlabels[nid]),list(lg2.nlabels[nid]))
                    #if there is some error
                    if cost > 0:
                            # add mismatch
                            nlabelMismatch = nlabelMismatch + cost
                            # add errors in error list
                            for (l1,l2) in errL:
                                    nodeconflicts = nodeconflicts + [ (nid, [ (l1, 1.0) ], [(l2, 1.0)] ) ]
                            # add node in error list
                            nodeClassError = nodeClassError.union([nid])

            # Two-sided comparison of *label sets* (look from absent edges in both
            # graphs!) Must check whether edge exists; '_' represents a "NONE"
            # label (no edge).

            # Identify the set of nodes with disagreeing edges.
            nodeEdgeError = set()
            for (graph,oGraph) in [ (self,lg2), (lg2,self) ]:
                    for npair in list(graph.elabels):
                            if not npair in oGraph.elabels \
                                            and (not graph.elabels[ npair ] == ['_']):
                                    (cost,errL) = self.cmpEdges(list(graph.elabels[ npair ]),['_'])
                                    elabelMismatch = elabelMismatch + cost

                                    (a,b) = npair
                                    
                                    # Record nodes in invalid edge
                                    nodeEdgeError.update([a,b])

                                    # DEBUG: Need to indicate correctly *which* graph has the
                                    # missing edge; this graph (1st) or the other (listed 2nd).
                                    if graph == self:
                                            for (l1,l2) in errL:
                                                    edgeconflicts.append((npair, [ (l1, 1.0) ], [(l2, 1.0)] ) )
                                            
                                    else:
                                            for (l1,l2) in errL:
                                                    edgeconflicts.append((npair, [ (l2, 1.0) ], [(l1, 1.0)] ) )

            # Obtain number of primitives with an error of any sort.
            nodeError = nodeClassError.union(nodeEdgeError)

            # One-sided comparison for common edges. Compared by cmpEdges
            for npair in list(self.elabels):
                    if npair in list(lg2.elabels):
                            (cost,errL) = self.cmpEdges(list(self.elabels[npair]),list(lg2.elabels[npair]))
                            if cost > 0:
                                    elabelMismatch = elabelMismatch + cost
                                    (a,b) = npair
                                    
                                    # Record nodes in invalid edge
                                    nodeEdgeError.update([a,b])
                                    for (l1,l2) in errL:
                                            edgeconflicts.append((npair, [ (l1, 1.0) ], [(l2, 1.0)] ) )

            # Now compute segmentation differences.
            (segEdgeMismatch, segDiffs, correctSegs, segmentMetrics, segRelDiffs) \
                            = self.compareSegments(lg2)

            # UNDIRECTED/NODE PAIR METRICS
            # Compute number of invalid nodePairs
            badPairs = {}
            for ((n1, n2), _, _) in edgeconflicts:
                    if not (n2, n1) in badPairs:
                            badPairs[(n1, n2)] = True
            incorrectPairs = len(badPairs)

            # Compute number of mis-segmented node pairs.
            badSegPairs = set([])
            for node in list(segDiffs):
                    for other in segDiffs[node][0]:
                            if node != other and (other, node) not in badSegPairs:
                                    badSegPairs.add((node, other))
                    for other in segDiffs[node][1]:
                            if  node != other and (other, node)not in badSegPairs:
                                    badSegPairs.add((node, other))
            segPairErrors = len(badSegPairs)

            # Compute performance metrics; avoid divisions by 0.
            cerror = ("D_C", nlabelMismatch) 
            lerror = ("D_L", elabelMismatch) 
            serror = ("D_S", segEdgeMismatch) 
            rerror = ("D_R", elabelMismatch - segEdgeMismatch)
            aerror = ("D_B", nlabelMismatch + elabelMismatch) 

            # DEBUG:
            # Delta E BASE CASE: for a single node, which is absent in the other
            # file, set label and segment edge mismatches to 1 (in order
            # to obtain 1.0 as the error metric, i.e. total error).
            if len(list(self.nlabels)) == 1 and \
                            (len(self.absentNodes) > 0 or \
                            len(lg2.absentNodes) > 0):
                    elabelMismatch = 1
                    segEdgeMismatch = 1
            
            errorVal = 0.0
            if numEdges > 0: