Remove depedencies to Kaffe + fix test bug

python/models/tensorflow_unet.py

@@ -1,21 +1,20 @@
 import numpy as np
 import tensorflow as tf
-from .VGG_ILSVRC_16_layers import VGG_ILSVRC_16_layers as VGG16_net
-
-
-###########################################
-## NEEDED TO USE MAXPOOL WITh ARGMAX
-###########################################
-from tensorflow.python.framework import ops
-from tensorflow.python.ops import gen_nn_ops
-@ops.RegisterGradient("MaxPoolWithArgmax")
-def _MaxPoolGradWithArgmax(op, grad, unused_argmax_grad):
-    return gen_nn_ops._max_pool_grad_with_argmax(op.inputs[0],
-                          grad,
-                          op.outputs[1],
-                          op.get_attr("ksize"),
-                          op.get_attr("strides"),
-                          padding=op.get_attr("padding"))
+
+
+# ###########################################
+# ## NEEDED TO USE MAXPOOL WITh ARGMAX
+# ###########################################
+# from tensorflow.python.framework import ops
+# from tensorflow.python.ops import gen_nn_ops
+# @ops.RegisterGradient("MaxPoolWithArgmax")
+# def _MaxPoolGradWithArgmax(op, grad, unused_argmax_grad):
+#     return gen_nn_ops._max_pool_grad_with_argmax(op.inputs[0],
+#                           grad,
+#                           op.outputs[1],
+#                           op.get_attr("ksize"),
+#                           op.get_attr("strides"),
+#                           padding=op.get_attr("padding"))
 
 def print_activations(t):
     print(t.op.name, ' ', t.get_shape().as_list())
@@ -27,7 +26,7 @@ def print_activations(t):
     padding="SAME",
     activation_fn=tf.nn.relu,
     #weights_initializer=tf.truncated_normal_initializer(0, 0.1)
-    weights_initializer=tf.contrib.layers.xavier_initializer()
+    weights_initializer=tf.contrib.layers.xavier_initializer(), **kwargs
     )
 convolution2d_transpose = lambda *args, **kwargs : tf.contrib.layers.convolution2d_transpose(*args,
     kernel_size=[3,3],
@@ -37,8 +36,56 @@ def print_activations(t):
     weights_initializer=tf.contrib.layers.xavier_initializer(), **kwargs
     )
 
+
+class VGG16_net:
+
+    def __init__(self, images):
+        self.layers = {}
+        self.layers["conv1_1"] = convolution2d(images, 64, stride=1, scope="conv1_1")
+        self.layers["conv1_2"] = convolution2d(self.layers["conv1_1"], 64, stride=1, scope="conv1_2")
+
+        self.layers["pool1"] = tf.nn.max_pool(self.layers["conv1_2"], [1, 2, 2, 1], [1,2,2,1], padding="SAME")
+
+        self.layers["conv2_1"] = convolution2d(self.layers["pool1"], 128, stride=1, scope="conv2_1")
+        self.layers["conv2_2"] = convolution2d(self.layers["conv2_1"], 128, stride=1, scope="conv2_2")
+
+        self.layers["pool2"] = tf.nn.max_pool(self.layers["conv2_2"], [1, 2, 2, 1], [1,2,2,1], padding="SAME")
+
+        self.layers["conv3_1"] = convolution2d(self.layers["pool2"], 256, stride=1, scope="conv3_1")
+        self.layers["conv3_2"] = convolution2d(self.layers["conv3_1"], 256, stride=1, scope="conv3_2")
+        self.layers["conv3_3"] = convolution2d(self.layers["conv3_2"], 256, stride=1, scope="conv3_3")
+
+        self.layers["pool3"] = tf.nn.max_pool(self.layers["conv3_3"], [1, 2, 2, 1], [1,2,2,1], padding="SAME")
+
+        self.layers["conv4_1"] = convolution2d(self.layers["pool3"], 512, stride=1, scope="conv4_1")
+        self.layers["conv4_2"] = convolution2d(self.layers["conv4_1"], 512, stride=1, scope="conv4_2")
+        self.layers["conv4_3"] = convolution2d(self.layers["conv4_2"], 512, stride=1, scope="conv4_3")
+
+        self.layers["pool4"] = tf.nn.max_pool(self.layers["conv4_3"], [1, 2, 2, 1], [1,2,2,1], padding="SAME")
+
+        self.layers["conv5_1"] = convolution2d(self.layers["pool4"], 512, stride=1, scope="conv5_1")
+        self.layers["conv5_2"] = convolution2d(self.layers["conv5_1"], 512, stride=1, scope="conv5_2")
+        self.layers["conv5_3"] = convolution2d(self.layers["conv5_2"], 512, stride=1, scope="conv5_3")
+
+    def load(self, filename, variable_scope, session):
+        with tf.variable_scope(variable_scope) as scope:
+            data_dict = np.load(filename, encoding="bytes").tolist()
+            for op_name in data_dict:
+                with tf.variable_scope(op_name, reuse=True):
+                    for param_name, data in data_dict[op_name].items():
+                        param_name = param_name.decode('utf-8')
+                        try:
+                            var = tf.get_variable(param_name)
+                            session.run(var.assign(data))
+                        except ValueError:
+                            print("Error", op_name, param_name)
+                            #if not ignore_missing:
+                            #    raise
+
+
+
 def model(images, label_nbr, is_training=None):
-    net = VGG16_net({'data': images})
+    net = VGG16_net(images)
 
     vgg_output = net.layers["conv5_3"]
 
@@ -69,7 +116,5 @@ def model(images, label_nbr, is_training=None):
     deconv_net.append(convolution2d_transpose(deconv_net[-1], 64, stride=1))
     deconv_net.append(convolution2d_transpose(deconv_net[-1], label_nbr, stride=1))
     print_activations(deconv_net[-1])
-
-
-
+
     return deconv_net, net

python/tensorflow_tester_backprojeter.py

@@ -163,7 +163,7 @@ def createLabelPLY(self, filename,
         save_dir):
 
         # create the semantizer
-        semantizer = Sem3D()
+        semantizer = PcTls.Semantic3D()
         semantizer.set_voxel_size(0.1)
 
         # loading data

python/tensorflow_trainer.py

@@ -57,7 +57,8 @@ def train(self,
 
             # load net weights if needed
             if net is not None:
-                net.load(net_weights_init, sess)
+                net.load(net_weights_init, variable_scope = variable_scope, session=sess)
+                #net.load(net_weights_init, sess)
 
             # create the list of images in the folder
             directory = os.path.join(dir_images, images_root)

sem3d_test_backproj_tf.py

@@ -24,20 +24,20 @@
 
 filenames = [
         "birdfountain_station1_xyz_intensity_rgb",
-        "castleblatten_station1_intensity_rgb",
-        "castleblatten_station5_xyz_intensity_rgb",
-        "marketplacefeldkirch_station1_intensity_rgb",
-        "marketplacefeldkirch_station4_intensity_rgb",
-        "marketplacefeldkirch_station7_intensity_rgb",
-        "sg27_station10_intensity_rgb",
-        "sg27_station3_intensity_rgb",
-        "sg27_station6_intensity_rgb",
-        "sg27_station8_intensity_rgb",
-        "sg28_station2_intensity_rgb",
-        "sg28_station5_xyz_intensity_rgb",
-        "stgallencathedral_station1_intensity_rgb",
-        "stgallencathedral_station3_intensity_rgb",
-        "stgallencathedral_station6_intensity_rgb"
+        # "castleblatten_station1_intensity_rgb",
+        # "castleblatten_station5_xyz_intensity_rgb",
+        # "marketplacefeldkirch_station1_intensity_rgb",
+        # "marketplacefeldkirch_station4_intensity_rgb",
+        # "marketplacefeldkirch_station7_intensity_rgb",
+        # "sg27_station10_intensity_rgb",
+        # "sg27_station3_intensity_rgb",
+        # "sg27_station6_intensity_rgb",
+        # "sg27_station8_intensity_rgb",
+        # "sg28_station2_intensity_rgb",
+        # "sg28_station5_xyz_intensity_rgb",
+        # "stgallencathedral_station1_intensity_rgb",
+        # "stgallencathedral_station3_intensity_rgb",
+        # "stgallencathedral_station6_intensity_rgb"
     ]