a better load_vgg

tensorpack · Apr 20, 2016 · bd0ca73 · bd0ca73
1 parent 5476b48
commit bd0ca73
Showing 1 changed file with 37 additions and 47 deletions.
diff --git a/examples/load_vgg16.py b/examples/load_vgg16.py
@@ -20,6 +20,12 @@
 from tensorpack.dataflow import *
 from tensorpack.dataflow.dataset import ILSVRCMeta
 
+"""
+Usage:
+    python2 -m tensorpack.utils.loadcaffe PATH/TO/models/VGG/{VGG_ILSVRC_16_layers_deploy.prototxt,VGG_ILSVRC_16_layers.caffemodel} vgg16.npy
+    ./load_vgg16.py --load vgg16.npy --input cat.png
+"""
+
 class Model(ModelDesc):
     def _get_input_vars(self):
         return [InputVar(tf.float32, (None, 224, 224, 3), 'input'),
@@ -31,33 +37,34 @@ def _get_cost(self, inputs, is_training):
 
         image, label = inputs
 
-        # 224
-        l = Conv2D('conv1_1', image, out_channel=64, kernel_shape=3)
-        l = Conv2D('conv1_2', l, out_channel=64, kernel_shape=3)
-        l = MaxPooling('pool1', l, 2, stride=2, padding='VALID')
-        # 112
-
-        l = Conv2D('conv2_1', l, out_channel=128, kernel_shape=3)
-        l = Conv2D('conv2_2', l, out_channel=128, kernel_shape=3)
-        l = MaxPooling('pool2', l, 2, stride=2, padding='VALID')
-        # 56
-
-        l = Conv2D('conv3_1', l, out_channel=256, kernel_shape=3)
-        l = Conv2D('conv3_2', l, out_channel=256, kernel_shape=3)
-        l = Conv2D('conv3_3', l, out_channel=256, kernel_shape=3)
-        l = MaxPooling('pool3', l, 2, stride=2, padding='VALID')
-        # 28
-
-        l = Conv2D('conv4_1', l, out_channel=512, kernel_shape=3)
-        l = Conv2D('conv4_2', l, out_channel=512, kernel_shape=3)
-        l = Conv2D('conv4_3', l, out_channel=512, kernel_shape=3)
-        l = MaxPooling('pool4', l, 2, stride=2, padding='VALID')
-        # 14
-
-        l = Conv2D('conv5_1', l, out_channel=512, kernel_shape=3)
-        l = Conv2D('conv5_2', l, out_channel=512, kernel_shape=3)
-        l = Conv2D('conv5_3', l, out_channel=512, kernel_shape=3)
-        l = MaxPooling('pool5', l, 2, stride=2, padding='VALID')
+        with argscope(Conv2D, kernel_shape=3):
+            # 224
+            l = Conv2D('conv1_1', image, 64)
+            l = Conv2D('conv1_2', l, 64)
+            l = MaxPooling('pool1', l, 2, stride=2, padding='VALID')
+            # 112
+
+            l = Conv2D('conv2_1', l, 128)
+            l = Conv2D('conv2_2', l, 128)
+            l = MaxPooling('pool2', l, 2, stride=2, padding='VALID')
+            # 56
+
+            l = Conv2D('conv3_1', l, 256)
+            l = Conv2D('conv3_2', l, 256)
+            l = Conv2D('conv3_3', l, 256)
+            l = MaxPooling('pool3', l, 2, stride=2, padding='VALID')
+            # 28
+
+            l = Conv2D('conv4_1', l, 512)
+            l = Conv2D('conv4_2', l, 512)
+            l = Conv2D('conv4_3', l, 512)
+            l = MaxPooling('pool4', l, 2, stride=2, padding='VALID')
+            # 14
+
+            l = Conv2D('conv5_1', l, 512)
+            l = Conv2D('conv5_2', l, 512)
+            l = Conv2D('conv5_3', l, 512)
+            l = MaxPooling('pool5', l, 2, stride=2, padding='VALID')
         # 7
 
         l = FullyConnected('fc6', l, 4096)
@@ -67,26 +74,9 @@ def _get_cost(self, inputs, is_training):
         logits = FullyConnected('fc8', l, out_dim=1000, nl=tf.identity)
         prob = tf.nn.softmax(logits, name='output')
 
-        y = one_hot(label, 1000)
-        cost = tf.nn.softmax_cross_entropy_with_logits(logits, y)
-        cost = tf.reduce_mean(cost, name='cross_entropy_loss')
-        tf.add_to_collection(MOVING_SUMMARY_VARS_KEY, cost)
-
-        # compute the number of failed samples, for ValidationError to use at test time
-        wrong = tf.not_equal(
-            tf.cast(tf.argmax(prob, 1), tf.int32), label)
-        wrong = tf.cast(wrong, tf.float32)
-        nr_wrong = tf.reduce_sum(wrong, name='wrong')
-        # monitor training error
-        tf.add_to_collection(
-            MOVING_SUMMARY_VARS_KEY, tf.reduce_mean(wrong, name='train_error'))
-
-        # weight decay on all W of fc layers
-        wd_cost = tf.mul(1e-4,
-                         regularize_cost('fc.*/W', tf.nn.l2_loss),
-                         name='regularize_loss')
-        tf.add_to_collection(MOVING_SUMMARY_VARS_KEY, wd_cost)
-        return tf.add_n([wd_cost, cost], name='cost')
+        cost = tf.nn.sparse_softmax_cross_entropy_with_logits(logits, label)
+        cost = tf.reduce_mean(cost, name='cost')
+        return cost
 
 def run_test(path, input):
     param_dict = np.load(path).item()