chrysanthème

.gitignore

@@ -3,3 +3,4 @@
 .cache/
 .settings/
 __pycache__/
+*.bak

LICENSE

@@ -1,6 +1,7 @@
 BSD 3-Clause License
 
-Copyright (c) 2016, Transkribus
+Copyright (c) 2016-2019, NAVER LABS Europe
+
 All rights reserved.
 
 Redistribution and use in source and binary forms, with or without

README.md

@@ -1,17 +1,24 @@
 # TranskribusDU
 Document Understanding tools
 
-### Requirements, installation & testing
+Updated: 2019-11-20
+
+### Requirements, installation
 
 #### Python
 
 * Install [Python] 3.x 
 
 We recommend installing __anaconda3__ . You can then train using pystruct and/or tensorflow (both to be installed on top of anaconda).
 
- * conda install shapely rtree
+ * conda install shapely rtree lxml scipy
+ * pip install future scikit-learn pytest --upgrade
+
+To learn with pystruct (using a graph-CRF model):
+ * pip install cvxopt ad3 pystruct --upgrade
+
+To learn with Tensorflow (using an Edge Convolutional Network):
  * conda install -c anaconda tensorflow(-gpu)
- * pip install future lxml scipy scikit-learn pytest cvxopt ad3 pystruct --upgrade
 
 ### Usage
  * see use-cases

RELEASE_NOTES.txt

@@ -2,11 +2,36 @@
 		RELEASE NOTES - TranskribusDU
 		-----------------------------
 
---- Edelweiss - 2019-07-22
-- ECN/GAT
-- conjugate
-- --g1 --g2
-- table understanding
+--- Chrysanthème - 2019-11-21
+- ICDAR19 papers are reproducible
+- major code reorganisation
+- Multipage XML bug fixes
+- standard projection profile method
+- convex hull for cluster Coords
+- ECN ensemble bug fix
+- various bug fixes
+- --server mode
+- segmentation task using agglomerative clustering
+- Json input
+- pipe example 
+- table reconstruction
+- generic features (when no page info)
+- edge features reworked
+- cluster evaluation metrics
+
+
+--- Iris - 2019-04-25
+- CRF, ECN GAT supported
+- conjugate mode supported
+- --vld option to specify a validation set, or a ratio of validation graphs 
+	taken from the training set. The best model on validation set is kept.
+- --graph option to store the edges in the output XML
+- --max_iter applies to all learning methods
+- --seed to seed the randomizer with a constant
+- dynamic load of the learners
+- major code re-organization
+- for example of use, see in tasks: DU_TABLE_BIO.py or DU_Table_Row_Edge.py
+
 
 --- Jonquille - 2017-04-28
 - multi-type classification supported
@@ -30,21 +55,10 @@
 
 
 ---------------------------------------------------------------------------------
-    Copyright (C) 2016, 2017 H. Déjean, JL Meunier
-
-    This program is free software: you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation, either version 3 of the License, or
-    (at your option) any later version.
+    Copyright (C)  2016-2019  H. Déjean, JL Meunier
 
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
 
-    You should have received a copy of the GNU General Public License
-    along with this program.  If not, see <http://www.gnu.org/licenses/>.
 
     Developed  for the EU project READ. The READ project has received funding 
     from the European Union's Horizon 2020 research and innovation programme 
-    under grant agreement No 674943.
+    under grant agreement No 674943.

TranskribusDU/ObjectModel/XMLDSBASELINEClass.py

@@ -45,12 +45,10 @@ def computePoints(self):
         if self.lPoints is None:
             self.lPoints = self.getAttribute('blpoints')
 #             print 'after split?',self.lPoints
-
-        self.lPoints = self.lPoints.replace(" ",",")
         if self.lPoints is not None:
             lX = list(map(lambda x:float(x),self.lPoints.split(',')))[0::2]
             lY = list(map(lambda x:float(x),self.lPoints.split(',')))[1::2]
-            self.lPoints = list(zip(lX,lY))
+            self.lPoints = zip(lX,lY)
 #             lY.sort()
 #             if len(lY)> 10:  ## if basline automatically generated: beg and end noisy 
 #                 lY= lY[1:-2]
@@ -67,9 +65,14 @@ def computePoints(self):
             import numpy as np
             a,b = np.polyfit(lX, lY, 1)
             self.setAngle(a)
-            self.setBx(b)
 #             ymax = a * self.getX2() +b
 #             ymin = a*self.getX() + b
+#             import libxml2
+#             verticalSep  = libxml2.newNode('PAGEBORDER')
+#             verticalSep.setProp('points', '%f,%f,%f,%f'%(self.getX(),ymin,self.getX2(),ymax))         
+# #             print 'p',self.getParent()
+# #             print 'pp',self.getParent().getParent() 
+#             self.getParent().getNode().addChild(verticalSep)
 
     """
         TO simulate 'DS' objects

TranskribusDU/ObjectModel/XMLDSCELLClass.py

@@ -8,18 +8,7 @@
     a class for table cell from a XMLDocument
     READ project 
 
-    This program is free software: you can redistribute it and/or modify
-    it under the terms of the GNU General Public License as published by
-    the Free Software Foundation, either version 3 of the License, or
-    (at your option) any later version.
 
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU General Public License for more details.
-
-    You should have received a copy of the GNU General Public License
-    along with this program.  If not, see <http://www.gnu.org/licenses/>.
     
     
     Developed  for the EU project READ. The READ project has received funding 
@@ -96,17 +85,7 @@ def fromDom(self,domNode):
         self.setNode(domNode)
         # get properties
         for prop in domNode.keys():
-            try:
-                self.addAttribute(prop,domNode.get(prop))
-                if prop =='x': self._x= float(domNode.get(prop))
-                elif prop =='y': self._y = float(domNode.get(prop))
-                elif prop =='height': self._h = float(domNode.get(prop))
-                elif prop =='width': self.setWidth(float(domNode.get(prop)))   
-            except:
-                self._x=-1
-                self._y=-1
-                self._h=0
-                self._w=0         
+            self.addAttribute(prop,domNode.get(prop))
 
         self.setIndex(int(self.getAttribute('row')),int(self.getAttribute('col')))
 

TranskribusDU/ObjectModel/XMLDSGRAHPLINEClass.py

@@ -40,10 +40,9 @@ def computePoints(self):
             self.lPoints = self.getAttribute('points')
 #             print 'after split?',self.lPoints
         if self.lPoints is not None:
-            lX=[float(x) for p in self.lPoints.split(' ') for x in p.split(',')[0::2]]
-#             lX = list(map(lambda x:float(x),self.lPoints.split(',')))[0::2]
-            lY = [float(x) for p in self.lPoints.split(' ') for x in p.split(',')[1::2]]
-            self.lPoints = list(zip(lX,lY))
+            lX = list(map(lambda x:float(x),self.lPoints.split(',')))[0::2]
+            lY = list(map(lambda x:float(x),self.lPoints.split(',')))[1::2]
+            self.lPoints = zip(lX,lY)
             try:
                 self.avgY = 1.0 * sum(lY)/len(lY)
             except ZeroDivisionError:
@@ -59,6 +58,10 @@ def computePoints(self):
 #             self.setAngle(a)
 #             ymax = a * self.getX2() +b
 #             ymin = a*self.getX() + b
+#             import libxml2
+#             verticalSep  = libxml2.newNode('PAGEBORDER')
+#             verticalSep.setProp('points', '%f,%f,%f,%f'%(self.getX(),ymin,self.getX2(),ymax))         
+#             self.getParent().getNode().addChild(verticalSep)
 
     """
         TO simulate 'DS' objects
@@ -76,7 +79,6 @@ def getWidth(self): return abs(float(self.getAttribute('width')))
 
 
     def setPoints(self,lp): self.lPoints = lp
-    def getPoints(self): return self.lPoints 
 
     def fromDom(self,domNode):
         """

TranskribusDU/ObjectModel/XMLDSLINEClass.py

@@ -36,16 +36,13 @@ def fromDom(self,domNode):
         self.setNode(domNode)
 
         # get properties
-#         for prop in domNode.keys():
-#             self.addAttribute(prop,domNode.get(prop))
-
         for prop in domNode.keys():
             self.addAttribute(prop,domNode.get(prop))
-            if prop =='x': self._x= float(domNode.get(prop))
-            elif prop =='y': self._y = float(domNode.get(prop))
-            elif prop =='height': self._h = float(domNode.get(prop))
-            elif prop =='width': self.setWidth(float(domNode.get(prop)))
-
+
+#         ctxt = domNode.doc.xpathNewContext()
+#         ctxt.setContextNode(domNode)
+#         ldomElts = ctxt.xpathEval('./%s'%(ds_xml.sTEXT))
+#         ctxt.xpathFreeContext()
         ldomElts = domNode.findall('./%s'%(ds_xml.sTEXT))
         for elt in ldomElts:
             myObject= XMLDSTEXTClass(elt)