Work on TF branch, 3

osmr · Oct 15, 2018 · 92e65dd · 92e65dd
1 parent 7f99c7a
commit 92e65dd
Show file tree

Hide file tree

Showing 4 changed files with 83 additions and 57 deletions.
diff --git a/gluon/gluoncv2/models/resnet.py b/gluon/gluoncv2/models/resnet.py
@@ -959,38 +959,38 @@ def _test():
     import numpy as np
     import mxnet as mx
 
-    pretrained = True
+    pretrained = False
 
     models = [
-        resnet10,
-        resnet12,
-        resnet14,
-        resnet16,
-        resnet18_wd4,
-        resnet18_wd2,
-        resnet18_w3d4,
+        # resnet10,
+        # resnet12,
+        # resnet14,
+        # resnet16,
+        # resnet18_wd4,
+        # resnet18_wd2,
+        # resnet18_w3d4,
 
         resnet18,
-        resnet34,
-        resnet50,
-        resnet50b,
-        resnet101,
-        resnet101b,
-        resnet152,
-        resnet152b,
-        # resnet200,
-        # resnet200b,
-
-        # seresnet18,
-        # seresnet34,
-        seresnet50,
-        # seresnet50b,
-        seresnet101,
-        # seresnet101b,
-        seresnet152,
-        # seresnet152b,
-        # seresnet200,
-        # seresnet200b,
+        # resnet34,
+        # resnet50,
+        # resnet50b,
+        # resnet101,
+        # resnet101b,
+        # resnet152,
+        # resnet152b,
+        # # resnet200,
+        # # resnet200b,
+        #
+        # # seresnet18,
+        # # seresnet34,
+        # seresnet50,
+        # # seresnet50b,
+        # seresnet101,
+        # # seresnet101b,
+        # seresnet152,
+        # # seresnet152b,
+        # # seresnet200,
+        # # seresnet200b,
     ]
 
     for model in models:

diff --git a/tensorflow_/models/common.py b/tensorflow_/models/common.py
@@ -72,6 +72,29 @@ def conv2d(x,
         use_bias=use_bias,
         name=name)
 
+    # import numpy as np
+    # if tf_padding == 'same':
+    #     tf_padding = 'SAME'
+    # elif tf_padding == 'valid':
+    #     tf_padding = 'VALID'
+    # else:
+    #     raise ValueError('Invalid padding: ' + str(tf_padding))
+    # x = tf.nn.convolution(
+    #     input=x,
+    #     filter=tf.Variable(np.zeros(kernel_size + (int(x.shape[1]), out_channels), np.float32),
+    #                        dtype=tf.float32, name=name + "/kernel"),
+    #     dilation_rate=(1, 1),
+    #     strides=strides,
+    #     padding=tf_padding,
+    #     # name=name + "/kernel1",
+    #     data_format='NCHW')
+    # if use_bias:
+    #     x = tf.nn.bias_add(
+    #         value=x,
+    #         bias=tf.Variable(np.zeros((out_channels,), np.float32), dtype=tf.float32, name=name + "/bias"),
+    #         data_format='NCHW')
+    #         # name=name + "/bias1")
+
     return x
 
 

diff --git a/tensorflow_/models/model_store.py b/tensorflow_/models/model_store.py
@@ -220,6 +220,8 @@ def load_model(sess,
             sess.run(dst_params[src_key].assign(src_params[src_key]))
         elif not ignore_extra:
             raise Exception("The file `{}` is incompatible with the model".format(file_path))
+        else:
+            print("Key `{}` is ignored".format(src_key))
 
 
 def download_model(sess,

diff --git a/tensorflow_/models/resnet.py b/tensorflow_/models/resnet.py
@@ -999,40 +999,40 @@ def _test():
     pretrained = False
 
     models = [
-        resnet10,
-        resnet12,
-        resnet14,
-        resnet16,
-        resnet18_wd4,
-        resnet18_wd2,
-        resnet18_w3d4,
+        # resnet10,
+        # resnet12,
+        # resnet14,
+        # resnet16,
+        # resnet18_wd4,
+        # resnet18_wd2,
+        # resnet18_w3d4,
 
         resnet18,
-        resnet34,
-        resnet50,
-        resnet50b,
-        resnet101,
-        resnet101b,
-        resnet152,
-        resnet152b,
-        resnet200,
-        resnet200b,
-
-        seresnet18,
-        seresnet34,
-        seresnet50,
-        seresnet50b,
-        seresnet101,
-        seresnet101b,
-        seresnet152,
-        seresnet152b,
-        seresnet200,
-        seresnet200b,
+        # resnet34,
+        # resnet50,
+        # resnet50b,
+        # resnet101,
+        # resnet101b,
+        # resnet152,
+        # resnet152b,
+        # resnet200,
+        # resnet200b,
+        #
+        # seresnet18,
+        # seresnet34,
+        # seresnet50,
+        # seresnet50b,
+        # seresnet101,
+        # seresnet101b,
+        # seresnet152,
+        # seresnet152b,
+        # seresnet200,
+        # seresnet200b,
     ]
 
     for model in models:
 
-        net, _ = model(pretrained=pretrained)
+        net = model(pretrained=pretrained)
 
         x = tf.placeholder(
             dtype=tf.float32,
@@ -1072,7 +1072,8 @@ def _test():
 
         with tf.Session() as sess:
             sess.run(tf.global_variables_initializer())
-            y = sess.run(y_net, feed_dict={x: np.zeros((1, 3, 224, 224), np.float32)})
+            x_value = np.zeros((1, 3, 224, 224), np.float32)
+            y = sess.run(y_net, feed_dict={x: x_value})
             assert (y.shape == (1, 1000))
         tf.reset_default_graph()