Owang dev

core/evaluator_interactive.py

@@ -1,4 +1,6 @@
 import tensorflow as tf
+import matplotlib
+matplotlib.use('agg')
 import matplotlib.pyplot as plt
 from matplotlib.widgets import Slider
 import numpy as np
@@ -38,15 +40,15 @@ def construct_graph(eval_config):
         depth_320, depth_net = build_depth_resnet.build(image_320_tensor, is_training)
         focal_depth = depth_320[0, focal_y_tensor, focal_x_tensor, 0]
         depth_320_signed = (depth_320 - focal_depth) * aperture_tensor
-        pre_dof_320, lensblur_net, feature_net = build_lensblur.build(image_320_tensor, depth_320_signed, is_training) 
+        pre_dof_320, lensblur_net, feature_net = build_lensblur.build(image_320_tensor, depth_320_signed, is_training)
         pre_dof_640, sr_net = build_sr.build(image_320_tensor, depth_320_signed, pre_dof_320, image_640_tensor, is_training)
         shape_640 = tf.shape(pre_dof_640)
         depth_640_signed = tf.image.resize_bilinear(depth_320_signed, [shape_640[1], shape_640[2]])
-        pre_dof_1280, sr_net = build_sr.build(image_640_tensor, depth_640_signed, pre_dof_640, image_1280_tensor, is_training) 
+        pre_dof_1280, sr_net = build_sr.build(image_640_tensor, depth_640_signed, pre_dof_640, image_1280_tensor, is_training)
 
 
     variables_to_restore = tf.global_variables()
-    
+
     session_config = tf.ConfigProto()
     session_config.gpu_options.per_process_gpu_memory_fraction = 0.8
     sess = tf.Session(config=session_config)
@@ -65,7 +67,7 @@ def construct_graph(eval_config):
     else:
         raise ValueError("No pre-trained models")
 
-    
+
     def pre_dof(im):
         focal_x, focal_y = im.get_focal_point()
         aperture = im.aperture
@@ -105,6 +107,7 @@ def evaluate(eval_config):
     depth_handle = plt.imshow(pre_depth[0,:,:,0])
     renderdof = RenderDoF(im_handle, depth_handle, slider, im, run_dof_func, run_depth_func, start_id, image_names)
     plt.show()
+    #plt.savefig('output.jpg')
 
 def _normalize(x):
     x = x.astype(np.float32)
@@ -121,13 +124,13 @@ def __init__(self, im, depth=None, dof=None, x=0, y=0, aperture=1.0):
             self.scale = 4
             self.im_1280 = im
             #self.im_1280 = cv2.resize(self.im, (self.im.shape[1]/2, self.im.shape[0]/2), interpolation=cv2.INTER_AREA)
-            self.im_640 = cv2.resize(self.im_1280, (self.im_1280.shape[1]/2, self.im_1280.shape[0]/2), interpolation=cv2.INTER_AREA)
-            self.im_320 = cv2.resize(self.im_640, (self.im_640.shape[1]/2, self.im_640.shape[0]/2), interpolation=cv2.INTER_AREA)
+            self.im_640 = cv2.resize(self.im_1280, (int(self.im_1280.shape[1]/2), int(self.im_1280.shape[0]/2)), interpolation=cv2.INTER_AREA)
+            self.im_320 = cv2.resize(self.im_640, (int(self.im_640.shape[1]/2),int(self.im_640.shape[0]/2)), interpolation=cv2.INTER_AREA)
         elif np.max(im.shape) > 640:
             self.scale = 2
             self.im_1280 = None
             self.im_640 = im
-            self.im_320 = cv2.resize(self.im_640, (self.im_640.shape[1]/2, self.im_640.shape[0]/2), interpolation=cv2.INTER_AREA)
+            self.im_320 = cv2.resize(self.im_640, (int(self.im_640.shape[1]/2), int(self.im_640.shape[0]/2)), interpolation=cv2.INTER_AREA)
         else:
             self.scale = 1
             self.im_1280 = None
@@ -208,7 +211,7 @@ def on_key(self, event):
             im = im.astype(np.float32) / 255
             im = ImageWrapper(im, aperture=self.slider.val/10.0)
             self.im = im
-            print self.im_names[self.id]
+            print(self.im_names[self.id])
         except:
             start_id = self.id
             exit()

eval.py

@@ -1,5 +1,5 @@
 import tensorflow as tf
-
+from network.network import Network
 from google.protobuf import text_format
 from proto.eval_config_pb2 import EvalConfig
 from core import evaluator_interactive as evaluator

model/eval.config

@@ -1,4 +1,4 @@
-depth_ckpt: './model/model_depth/model.ckpt-346224'
+depth_ckpt: './model/model_depth/model-346224'
 lensblur_ckpt: 'model/model_lensblur/model-199950'
 res_path: './data/res/'
 image_path: "./data/imgs/"

network/feature_net.py

@@ -1,4 +1,4 @@
-from network import Network
+from network.network import Network
 
 class FeaNet(Network):
     def __init__(self, input, is_training, trainable=True, bn_global=None, upsample_size=640):

network/kernel_net.py

@@ -1,4 +1,4 @@
-from network import Network
+from network.network import Network
 import tensorflow as tf
 
 class KernelNet(Network):
@@ -64,7 +64,7 @@ def setup(self):
         self.resize(shape=tf.shape(self.layers['image'])[1:3], name='upsample')
 
         #concat
-        (self.feed('upsample', 
+        (self.feed('upsample',
                    'bn1_shallow')
          .concat(3, name='concat'))
 

network/network.py

@@ -84,7 +84,7 @@ def feed(self, *args):
         assert len(args) != 0
         self.terminals = []
         for fed_layer in args:
-            if isinstance(fed_layer, basestring):
+            if isinstance(fed_layer, str):
                 try:
                     fed_layer = self.layers[fed_layer]
                 except KeyError:
@@ -102,7 +102,7 @@ def get_unique_name(self, prefix):
         '''
         ident = sum(t.startswith(prefix) for t, _ in self.layers.items()) + 1
         return '%s_%d' % (prefix, ident)
-    
+
     def make_var(self, name, shape, weight_decay=0, trainable=None, initializer=None):
         '''Creates a new TensorFlow variable.'''
         if trainable == None:
@@ -139,7 +139,7 @@ def conv(self,
         # Verify that the padding is acceptable
         self.validate_padding(padding)
         # Get the number of channels in the input
-        c_i = input.get_shape()[-1]
+        c_i = input.get_shape().as_list()[-1]
         # Verify that the grouping parameter is valid
         assert c_i % group == 0
         assert c_o % group == 0
@@ -191,7 +191,7 @@ def separable_conv(self,
         # Convolution for a given input and kernel
         if dilation_rate > 1:
             assert s_h == 1 and s_w == 1
-        separable_convolve = lambda i, k_depth, k_point: tf.nn.separable_conv2d(i, k_depth, k_point, strides=[1, s_h, s_w, 1], 
+        separable_convolve = lambda i, k_depth, k_point: tf.nn.separable_conv2d(i, k_depth, k_point, strides=[1, s_h, s_w, 1],
                 rate=[dilation_rate, dilation_rate], padding=padding, name=name)
         with tf.variable_scope(name) as scope:
             #kernel = self.make_var('weights', shape=[k_h, k_w, c_i / group, c_o])
@@ -302,7 +302,7 @@ def add(self, inputs, name):
     @layer
     def fc(self, input, num_out, name, relu=True):
         with tf.variable_scope(name) as scope:
-            input_shape = input.get_shape()
+            input_shape = input.get_shape().as_list()
             if input_shape.ndims == 4:
                 # The input is spatial. Vectorize it first.
                 dim = 1

network/resnet50_lrelu.py

@@ -1,4 +1,4 @@
-from network import Network
+from network.network import Network
 import tensorflow as tf
 
 class ResNet50(Network):
@@ -22,7 +22,7 @@ def setup(self):
              .conv(1, 1, 256, 1, 1, 1, biased=False, relu=False, name='res2a_branch2c')
              .batch_normalization(name='bn2a_branch2c'))
 
-        (self.feed('bn2a_branch1', 
+        (self.feed('bn2a_branch1',
                    'bn2a_branch2c')
              .add(name='res2a')
              .relu(name='res2a_relu')
@@ -33,7 +33,7 @@ def setup(self):
              .conv(1, 1, 256, 1, 1, 1, biased=False, relu=False, name='res2b_branch2c')
              .batch_normalization(name='bn2b_branch2c'))
 
-        (self.feed('res2a_relu', 
+        (self.feed('res2a_relu',
                    'bn2b_branch2c')
              .add(name='res2b')
              .relu(name='res2b_relu')
@@ -44,7 +44,7 @@ def setup(self):
              .conv(1, 1, 256, 1, 1, 1, biased=False, relu=False, name='res2c_branch2c')
              .batch_normalization(name='bn2c_branch2c'))
 
-        (self.feed('res2b_relu', 
+        (self.feed('res2b_relu',
                    'bn2c_branch2c')
              .add(name='res2c')
              .relu(name='res2c_relu')
@@ -59,7 +59,7 @@ def setup(self):
              .conv(1, 1, 512, 1, 1, 1, biased=False, relu=False, name='res3a_branch2c')
              .batch_normalization(name='bn3a_branch2c'))
 
-        (self.feed('bn3a_branch1', 
+        (self.feed('bn3a_branch1',
                    'bn3a_branch2c')
              .add(name='res3a')
              .relu(name='res3a_relu')
@@ -70,7 +70,7 @@ def setup(self):
              .conv(1, 1, 512, 1, 1, 1, biased=False, relu=False, name='res3b_branch2c')
              .batch_normalization(name='bn3b_branch2c'))
 
-        (self.feed('res3a_relu', 
+        (self.feed('res3a_relu',
                    'bn3b_branch2c')
              .add(name='res3b')
              .relu(name='res3b_relu')
@@ -81,7 +81,7 @@ def setup(self):
              .conv(1, 1, 512, 1, 1, 1, biased=False, relu=False, name='res3c_branch2c')
              .batch_normalization(name='bn3c_branch2c'))
 
-        (self.feed('res3b_relu', 
+        (self.feed('res3b_relu',
                    'bn3c_branch2c')
              .add(name='res3c')
              .relu(name='res3c_relu')
@@ -92,7 +92,7 @@ def setup(self):
              .conv(1, 1, 512, 1, 1, 1, biased=False, relu=False, name='res3d_branch2c')
              .batch_normalization(name='bn3d_branch2c'))
 
-        (self.feed('res3c_relu', 
+        (self.feed('res3c_relu',
                    'bn3d_branch2c')
              .add(name='res3d')
              .relu(name='res3d_relu')
@@ -107,7 +107,7 @@ def setup(self):
              .conv(1, 1, 1024, 1, 1, 1, biased=False, relu=False, name='res4a_branch2c')
              .batch_normalization(name='bn4a_branch2c'))
 
-        (self.feed('bn4a_branch1', 
+        (self.feed('bn4a_branch1',
                    'bn4a_branch2c')
              .add(name='res4a')
              .relu(name='res4a_relu')
@@ -118,7 +118,7 @@ def setup(self):
              .conv(1, 1, 1024, 1, 1, 1, biased=False, relu=False, name='res4b_branch2c')
              .batch_normalization(name='bn4b_branch2c'))
 
-        (self.feed('res4a_relu', 
+        (self.feed('res4a_relu',
                    'bn4b_branch2c')
              .add(name='res4b')
              .relu(name='res4b_relu')
@@ -129,7 +129,7 @@ def setup(self):
              .conv(1, 1, 1024, 1, 1, 1, biased=False, relu=False, name='res4c_branch2c')
              .batch_normalization(name='bn4c_branch2c'))
 
-        (self.feed('res4b_relu', 
+        (self.feed('res4b_relu',
                    'bn4c_branch2c')
              .add(name='res4c')
              .relu(name='res4c_relu')
@@ -140,7 +140,7 @@ def setup(self):
              .conv(1, 1, 1024, 1, 1, 1, biased=False, relu=False, name='res4d_branch2c')
              .batch_normalization(name='bn4d_branch2c'))
 
-        (self.feed('res4c_relu', 
+        (self.feed('res4c_relu',
                    'bn4d_branch2c')
              .add(name='res4d')
              .relu(name='res4d_relu')
@@ -151,7 +151,7 @@ def setup(self):
              .conv(1, 1, 1024, 1, 1, 1, biased=False, relu=False, name='res4e_branch2c')
              .batch_normalization(name='bn4e_branch2c'))
 
-        (self.feed('res4d_relu', 
+        (self.feed('res4d_relu',
                    'bn4e_branch2c')
              .add(name='res4e')
              .relu(name='res4e_relu')
@@ -161,7 +161,7 @@ def setup(self):
              .batch_normalization(relu=True, name='bn4f_branch2b')
              .conv(1, 1, 1024, 1, 1, 1, biased=False, relu=False, name='res4f_branch2c')
              .batch_normalization(name='bn4f_branch2c'))
-        (self.feed('res4e_relu', 
+        (self.feed('res4e_relu',
                    'bn4f_branch2c')
              .add(name='res4f')
              .relu(name='res4f_relu'))

network/srnet.py

@@ -1,4 +1,4 @@
-from network import Network
+from network.network import Network
 import tensorflow as tf
 
 class SRNet(Network):