Merge branch 'develop' into patchesselection

include/otbTensorflowMultisourceModelBase.h

@@ -50,14 +50,14 @@ namespace otb
  * Target nodes names of the TensorFlow graph that must be triggered can be set
  * with the SetTargetNodesNames.
  *
- * The OutputTensorNames consists in a strd::vector of std::string, and
+ * The OutputTensorNames consists in a std::vector of std::string, and
  * corresponds to the names of tensors that will be computed during the session.
  * As for input placeholders, output tensors field of expression
  * (OutputExpressionFields, a std::vector of SizeType), i.e. the output
  * space that the TensorFlow model will "generate", must be provided.
  *
  * Finally, a list of scalar placeholders can be fed in the form of std::vector
- * of std::string, each one expressing the assigment of a signle valued
+ * of std::string, each one expressing the assignment of a single valued
  * placeholder, e.g. "drop_rate=0.5 learning_rate=0.002 toto=true".
  * See otb::tf::ExpressionToTensor() to know more about syntax.
  *

include/otbTensorflowMultisourceModelLearningBase.h

@@ -23,7 +23,15 @@ namespace otb
 
 /**
  * \class TensorflowMultisourceModelLearningBase
- * \brief This filter is the base class for all learning filters.
+ * \brief This filter is the base class for all filters that input patches images.
+ *
+ * One input patches image consist in an image of size (pszx, pszy*n, nbands) where:
+ * -pszx   : is the width of one patch
+ * -pszy   : is the height of one patch
+ * -n      : is the number of patches in the patches image
+ * -nbands : is the number of channels in the patches image
+ *
+ * This filter verify that every patches images are consistent.
  *
  * The batch size can be set using the SetBatchSize() method.
  * The streaming can be activated to allow the processing of huge datasets.

include/otbTensorflowSampler.h

@@ -21,6 +21,9 @@
 // TF common
 #include "otbTensorflowCommon.h"
 
+// Tree iterator
+#include "itkPreOrderTreeIterator.h"
+
 namespace otb
 {
 

include/otbTensorflowStreamerFilter.h

@@ -11,8 +11,8 @@
 #ifndef otbTensorflowStreamerFilter_h
 #define otbTensorflowStreamerFilter_h
 
-// Image2image
 #include "itkImageToImageFilter.h"
+#include "itkProgressReporter.h"
 
 namespace otb
 {

python/create_savedmodel_pxs_fcn.py

+from tricks import *
+import sys
+import os
+
+nclasses=8
+
+def myModel(x1,x2):
+  
+  # The XS branch (input patches: 8x8x4)
+  conv1_x1 = tf.layers.conv2d(inputs=x1, filters=16, kernel_size=[5,5], padding="valid",
+                              activation=tf.nn.relu) # out size: 4x4x16
+  conv2_x1 = tf.layers.conv2d(inputs=conv1_x1, filters=32, kernel_size=[3,3], padding="valid",
+                              activation=tf.nn.relu) # out size: 2x2x32
+  conv3_x1 = tf.layers.conv2d(inputs=conv2_x1, filters=64, kernel_size=[2,2], padding="valid",
+                              activation=tf.nn.relu) # out size: 1x1x64
+  
+  # The PAN branch (input patches: 32x32x1)
+  conv1_x2 = tf.layers.conv2d(inputs=x2, filters=16, kernel_size=[5,5], padding="valid",
+                              activation=tf.nn.relu) # out size: 28x28x16
+  pool1_x2 = tf.layers.max_pooling2d(inputs=conv1_x2, pool_size=[2, 2], 
+                              strides=2) # out size: 14x14x16
+  conv2_x2 = tf.layers.conv2d(inputs=pool1_x2, filters=32, kernel_size=[5,5], padding="valid",
+                              activation=tf.nn.relu) # out size: 10x10x32
+  pool2_x2 = tf.layers.max_pooling2d(inputs=conv2_x2, pool_size=[2, 2],
+                              strides=2) # out size: 5x5x32
+  conv3_x2 = tf.layers.conv2d(inputs=pool2_x2, filters=64, kernel_size=[3,3], padding="valid",
+                              activation=tf.nn.relu) # out size: 3x3x64
+  conv4_x2 = tf.layers.conv2d(inputs=conv3_x2, filters=64, kernel_size=[3,3], padding="valid",
+                              activation=tf.nn.relu) # out size: 1x1x64
+  
+  # Stack features
+  features = tf.reshape(tf.stack([conv3_x1, conv4_x2], axis=3), 
+                        shape=[-1, 128], name="features")
+  
+  # 8 neurons for 8 classes
+  estimated = tf.layers.dense(inputs=features, units=nclasses, activation=None)
+  estimated_label = tf.argmax(estimated, 1, name="prediction")
+  
+  return estimated, estimated_label
+ 
+""" Main """
+# check number of arguments
+if len(sys.argv) != 2:
+  print("Usage : <output directory for SavedModel>")
+  sys.exit(1)
+
+# Create the graph
+with tf.Graph().as_default():
+  
+  # Placeholders
+  x1 = tf.placeholder(tf.float32, [None, None, None, 4], name="x1")
+  x2 = tf.placeholder(tf.float32, [None, None, None, 1], name="x2")
+  y  = tf.placeholder(tf.int32  , [None, None, None, 1], name="y")
+  lr = tf.placeholder_with_default(tf.constant(0.0002, dtype=tf.float32, shape=[]), 
+                                   shape=[], name="lr")
+  
+  # Output
+  y_estimated, y_label = myModel(x1,x2)
+  
+  # Loss function
+  cost = tf.losses.sparse_softmax_cross_entropy(labels=tf.reshape(y, [-1, 1]), 
+                                                logits=tf.reshape(y_estimated, [-1, nclasses]))
+  
+  # Optimizer
+  optimizer = tf.train.AdamOptimizer(learning_rate=lr, name="optimizer").minimize(cost)
+  
+  # Initializer, saver, session
+  init = tf.global_variables_initializer()
+  saver = tf.train.Saver( max_to_keep=20 )
+  sess = tf.Session()
+  sess.run(init)
+
+  # Create a SavedModel
+  CreateSavedModel(sess, ["x1:0", "x2:0", "y:0"], ["features:0", "prediction:0"], sys.argv[1])

python/create_savedmodel_simple_cnn.py

+from tricks import *
+import sys
+import os
+
+nclasses=8
+
+def myModel(x):
+  
+  # input patches: 16x16x4
+  conv1 = tf.layers.conv2d(inputs=x, filters=16, kernel_size=[5,5], padding="valid", 
+                           activation=tf.nn.relu) # out size: 12x12x16
+  pool1 = tf.layers.max_pooling2d(inputs=conv1, pool_size=[2, 2], strides=2) # out: 6x6x16
+  conv2 = tf.layers.conv2d(inputs=pool1, filters=16, kernel_size=[3,3], padding="valid", 
+                           activation=tf.nn.relu) # out size: 4x4x16
+  pool2 = tf.layers.max_pooling2d(inputs=conv2, pool_size=[2, 2], strides=2) # out: 2x2x16
+  conv3 = tf.layers.conv2d(inputs=pool2, filters=32, kernel_size=[2,2], padding="valid",
+                           activation=tf.nn.relu) # out size: 1x1x32
+  
+  # Features
+  features = tf.reshape(conv3, shape=[-1, 32], name="features")
+  
+  # 8 neurons for 8 classes
+  estimated = tf.layers.dense(inputs=features, units=nclasses, activation=None)
+  estimated_label = tf.argmax(estimated, 1, name="prediction")
+
+  return estimated, estimated_label
+ 
+""" Main """
+if len(sys.argv) != 2:
+  print("Usage : <output directory for SavedModel>")
+  sys.exit(1)
+
+# Create the TensorFlow graph
+with tf.Graph().as_default():
+  
+  # Placeholders
+  x = tf.placeholder(tf.float32, [None, None, None, 4], name="x")
+  y = tf.placeholder(tf.int32  , [None, None, None, 1], name="y")
+  lr = tf.placeholder_with_default(tf.constant(0.0002, dtype=tf.float32, shape=[]),
+                                   shape=[], name="lr")
+  
+  # Output
+  y_estimated, y_label = myModel(x)
+  
+  # Loss function
+  cost = tf.losses.sparse_softmax_cross_entropy(labels=tf.reshape(y, [-1, 1]), 
+                                                logits=tf.reshape(y_estimated, [-1, nclasses]))
+  
+  # Optimizer
+  optimizer = tf.train.AdamOptimizer(learning_rate=lr, name="optimizer").minimize(cost)
+  
+  # Initializer, saver, session
+  init = tf.global_variables_initializer()
+  saver = tf.train.Saver( max_to_keep=20 )
+  sess = tf.Session()
+  sess.run(init)
+
+  # Create a SavedModel
+  CreateSavedModel(sess, ["x:0", "y:0"], ["features:0", "prediction:0"], sys.argv[1])

python/create_savedmodel_simple_fcn.py

+from tricks import *
+import sys
+import os
+
+nclasses=8
+
+def myModel(x):
+  
+  # input patches: 16x16x4
+  conv1 = tf.layers.conv2d(inputs=x, filters=16, kernel_size=[5,5], padding="valid", 
+                           activation=tf.nn.relu) # out size: 12x12x16
+  conv2 = tf.layers.conv2d(inputs=conv1, filters=16, kernel_size=[5,5], padding="valid", 
+                           activation=tf.nn.relu) # out size: 8x8x16
+  conv3 = tf.layers.conv2d(inputs=conv2, filters=32, kernel_size=[5,5], padding="valid",
+                           activation=tf.nn.relu) # out size: 4x4x32
+  conv4 = tf.layers.conv2d(inputs=conv3, filters=32, kernel_size=[4,4], padding="valid",
+                           activation=tf.nn.relu) # out size: 1x1x32
+  
+  # Features
+  features = tf.reshape(conv4, shape=[-1, 32], name="features")
+  
+  # 8 neurons for 8 classes
+  estimated = tf.layers.dense(inputs=features, units=nclasses, activation=None)
+  estimated_label = tf.argmax(estimated, 1, name="prediction")
+
+  return estimated, estimated_label
+
+""" Main """
+if len(sys.argv) != 2:
+  print("Usage : <output directory for SavedModel>")
+  sys.exit(1)
+
+# Create the TensorFlow graph
+with tf.Graph().as_default():
+  
+  # Placeholders
+  x = tf.placeholder(tf.float32, [None, None, None, 4], name="x")
+  y = tf.placeholder(tf.int32  , [None, None, None, 1], name="y")
+  lr = tf.placeholder_with_default(tf.constant(0.0002, dtype=tf.float32, shape=[]),
+                                   shape=[], name="lr")
+  
+  # Output
+  y_estimated, y_label = myModel(x)
+  
+  # Loss function
+  cost = tf.losses.sparse_softmax_cross_entropy(labels=tf.reshape(y, [-1, 1]), 
+                                                logits=tf.reshape(y_estimated, [-1, nclasses]))
+  
+  # Optimizer
+  optimizer = tf.train.AdamOptimizer(learning_rate=lr, name="optimizer").minimize(cost)
+  
+  # Initializer, saver, session
+  init = tf.global_variables_initializer()
+  saver = tf.train.Saver( max_to_keep=20 )
+  sess = tf.Session()
+  sess.run(init)
+
+  # Create a SavedModel
+  CreateSavedModel(sess, ["x:0", "y:0"], ["features:0", "prediction:0"], sys.argv[1])

test/CMakeLists.txt

+otb_module_test()
+
+# Directories
+set(DATADIR ${CMAKE_CURRENT_SOURCE_DIR}/data)
+set(MODELSDIR ${CMAKE_CURRENT_SOURCE_DIR}/models)
+
+# Input images
+set(IMAGEXS ${DATADIR}/xs_subset.tif)
+set(IMAGEPAN ${DATADIR}/pan_subset.tif)
+set(IMAGEPXS ${DATADIR}/pxs_subset.tif)
+
+# Input models
+set(MODEL1 ${MODELSDIR}/model1)
+set(MODEL2 ${MODELSDIR}/model2)
+set(MODEL3 ${MODELSDIR}/model3)
+
+# Output images and baselines
+set(MODEL1_PB_OUT apTvClTensorflowModelServeCNN16x16PB.tif)
+set(MODEL2_PB_OUT apTvClTensorflowModelServeCNN8x8_32x32PB.tif)
+set(MODEL2_FC_OUT apTvClTensorflowModelServeCNN8x8_32x32FC.tif)
+set(MODEL3_PB_OUT apTvClTensorflowModelServeFCNN16x16PB.tif)
+set(MODEL3_FC_OUT apTvClTensorflowModelServeFCNN16x16FC.tif)
+
+#----------- Model serving : 1-branch CNN (16x16) Patch-Based ----------------
+otb_test_application(NAME TensorflowModelServeCNN16x16PB
+  APP  TensorflowModelServe
+  OPTIONS -source1.il ${IMAGEPXS}
+  -source1.rfieldx 16 -source1.rfieldy 16 -source1.placeholder x
+  -model.dir ${MODEL1} -output.names prediction
+  -out ${TEMP}/${MODEL1_PB_OUT}
+  VALID --compare-image ${EPSILON_6}
+  ${DATADIR}/${MODEL1_PB_OUT}
+  ${TEMP}/${MODEL1_PB_OUT})
+
+#----------- Model serving : 2-branch CNN (8x8, 32x32) Patch-Based ----------------
+otb_test_application(NAME apTvClTensorflowModelServeCNN8x8_32x32PB
+  APP  TensorflowModelServe
+  OPTIONS -source1.il ${IMAGEXS}
+  -source1.rfieldx 8 -source1.rfieldy 8 -source1.placeholder x1
+  -source2.il ${IMAGEPAN}
+  -source2.rfieldx 32 -source2.rfieldy 32 -source2.placeholder x2
+  -model.dir ${MODEL2} -output.names prediction
+  -out ${TEMP}/${MODEL2_PB_OUT}
+  VALID --compare-image ${EPSILON_6}
+  ${DATADIR}/${MODEL2_PB_OUT}
+  ${TEMP}/${MODEL2_PB_OUT})
+set_tests_properties(apTvClTensorflowModelServeCNN8x8_32x32PB PROPERTIES ENVIRONMENT "OTB_TF_NSOURCES=2;$ENV{OTB_TF_NSOURCES}")
+
+#----------- Model serving : 2-branch CNN (8x8, 32x32) Fully-Conv ----------------
+set(ENV{OTB_TF_NSOURCES} 2)
+otb_test_application(NAME apTvClTensorflowModelServeCNN8x8_32x32FC
+  APP  TensorflowModelServe
+  OPTIONS -source1.il ${IMAGEXS}
+  -source1.rfieldx 8 -source1.rfieldy 8 -source1.placeholder x1
+  -source2.il ${IMAGEPAN}
+  -source2.rfieldx 32 -source2.rfieldy 32 -source2.placeholder x2
+  -model.dir ${MODEL2} -output.names prediction -output.spcscale 4
+  -out ${TEMP}/${MODEL2_FC_OUT}
+  VALID --compare-image ${EPSILON_6}
+  ${DATADIR}/${MODEL2_FC_OUT}
+  ${TEMP}/${MODEL2_FC_OUT})
+set_tests_properties(apTvClTensorflowModelServeCNN8x8_32x32FC PROPERTIES ENVIRONMENT "OTB_TF_NSOURCES=2;$ENV{OTB_TF_NSOURCES}")
+
+#----------- Model serving : 1-branch FCNN (16x16) Patch-Based ----------------
+set(ENV{OTB_TF_NSOURCES} 1)
+otb_test_application(NAME apTvClTensorflowModelServeFCNN16x16PB
+  APP  TensorflowModelServe
+  OPTIONS -source1.il ${IMAGEPXS}
+  -source1.rfieldx 16 -source1.rfieldy 16 -source1.placeholder x
+  -model.dir ${MODEL3} -output.names prediction
+  -out ${TEMP}/${MODEL3_PB_OUT}
+  VALID --compare-image ${EPSILON_6}
+  ${DATADIR}/${MODEL3_PB_OUT}
+  ${TEMP}/${MODEL3_PB_OUT})
+
+#----------- Model serving : 1-branch FCNN (16x16) Fully-conv ----------------
+set(ENV{OTB_TF_NSOURCES} 1)
+otb_test_application(NAME apTvClTensorflowModelServeFCNN16x16FC
+  APP  TensorflowModelServe
+  OPTIONS -source1.il ${IMAGEPXS}
+  -source1.rfieldx 16 -source1.rfieldy 16 -source1.placeholder x
+  -model.dir ${MODEL3} -output.names prediction -model.fullyconv on
+  -out ${TEMP}/${MODEL3_FC_OUT}
+  VALID --compare-image ${EPSILON_6}
+  ${DATADIR}/${MODEL3_FC_OUT}
+  ${TEMP}/${MODEL3_FC_OUT})
+
+