diff options
Diffstat (limited to 'losses.py')
| -rw-r--r-- | losses.py | 25 |
1 files changed, 16 insertions, 9 deletions
@@ -68,9 +68,12 @@ class YourModel(tf.keras.Model): # for layer in self.vgg16: # layer.trainable = False - self.indexed_layers = [layer for layer in self.vgg16 if layer.name == "conv1"] + self.indexed_layers = [layer for layer in self.vgg16 if "conv1" in layer.name] print(self.indexed_layers) - self.desired = [layer.name for layer in self.vgg16 if layer.name == "conv1"] + self.desired = [layer.name for layer in self.vgg16 if "conv1" in layer.name] + + # create a map of the layers to their corresponding number of filters if it is a convolutional layer + self.layer_to_filters = {layer.name: layer.filters for layer in self.vgg16 if "conv" in layer.name} def call(self, x): layers = [] @@ -99,15 +102,18 @@ class YourModel(tf.keras.Model): return (self.alpha * content_l) + (self.beta * style_l) def content_loss(self, photo_layers, input_layers): + print(photo_layers, input_layers) L_content = tf.reduce_mean(tf.square(photo_layers - input_layers)) print(L_content) return L_content - def layer_loss(art_layers, input_layers, layer): + def layer_loss(self, art_layers, input_layers, layer): + # vectorize the art_layers + art_layers = tf.reshape(art_layers, (-1, art_layers.shape[-1])) + # vectorize the input_layers + input_layers = tf.reshape(input_layers, (-1, input_layers.shape[-1])) - #vectorize the inputs - art_vector = art_layers.reshape(-1, 224**2) - input_vector = input_layers.reshape(-1, 224**2) + print('layer', layer.name, self.layer_to_filters[layer.name]) # get the gram matrix input_dim = input_layers.shape[0] @@ -122,18 +128,19 @@ class YourModel(tf.keras.Model): # N depends on # of filters in the layer, M depends on hight and width of feature map M_l = art_layers.shape[0] * art_layers.shape[1] - # layer.filteres might not work - E_l = 1/4 * (layer.filters**(-2)) * (M_l**(-2)) * np.sum(np.square(G - input_layers)) + # layer.filters might not work + E_l = 1/4 * (self.layer_to_filters[layer.name]**(-2)) * (M_l**(-2)) * np.sum(np.square(G - input_layers)) # while Sotech is botty: # Jayson_tatum.tear_acl() # return ("this is just another day") + return E_l def style_loss(self, art_layers, input_layers): L_style = 0 for layer in self.indexed_layers: L_style += self.layer_loss(art_layers, input_layers, layer) - print('this is style loss',L_style) + print('this is style loss', L_style) return L_style def train_step(self): |