Example of how to train the lower-dimensional InfoGAN used in the exp…

…eriments.

examples/GAN/InfoGAN-mnist.py

@@ -32,11 +32,10 @@
 NUM_CLASS = 10
 NUM_UNIFORM = 2
 DIST_PARAM_DIM = NUM_CLASS + NUM_UNIFORM
-NOISE_DIM = 62
+NOISE_DIM = 3
 # prior: the assumption how the latent factors are presented in the dataset
 DIST_PRIOR_PARAM = [1.] * NUM_CLASS + [0.] * NUM_UNIFORM
 
-
 def shapeless_placeholder(x, axis, name):
     """
     Make the static shape of a tensor less specific.
@@ -108,6 +107,7 @@ class Model(GANModelDesc):
     def inputs(self):
         return [tf.TensorSpec((None, 28, 28), tf.float32, 'input')]
 
+    @auto_reuse_variable_scope
     def generator(self, z):
         l = FullyConnected('fc0', z, 1024, activation=BNReLU)
         l = FullyConnected('fc1', l, 128 * 7 * 7, activation=BNReLU)
@@ -146,6 +146,7 @@ def build_graph(self, real_sample):
         z_noise = shapeless_placeholder(
             tf.random_uniform([BATCH, NOISE_DIM], -1, 1), 0, name='z_noise')
         z = tf.concat([zc, z_noise], 1, name='z')
+        z2 = tf.placeholder('float', [None, int(z.shape[1])], name='z2')
 
         with argscope([Conv2D, Conv2DTranspose, FullyConnected],
                       kernel_initializer=tf.truncated_normal_initializer(stddev=0.02)):
@@ -154,6 +155,9 @@ def build_graph(self, real_sample):
                 fake_sample_viz = tf.cast((fake_sample) * 255.0, tf.uint8, name='viz')
                 tf.summary.image('gen', fake_sample_viz, max_outputs=30)
 
+                fake_sample2 = self.generator(z2)
+                fake_sample_viz2 = tf.identity(fake_sample2, name='viz2')
+
             # may need to investigate how bn stats should be updated across two discrim
             with tf.variable_scope('discrim'):
                 real_pred, _ = self.discriminator(real_sample)
@@ -220,8 +224,8 @@ def sample(model_path):
     pred = OfflinePredictor(PredictConfig(
         session_init=SmartInit(model_path),
         model=Model(),
-        input_names=['z_code', 'z_noise'],
-        output_names=['gen/viz']))
+        input_names=['z2'],
+        output_names=['gen/viz2']))
 
     # sample all one-hot encodings (10 times)
     z_cat = np.tile(np.eye(10), [10, 1])
@@ -231,6 +235,13 @@ def sample(model_path):
 
     IMG_SIZE = 400
 
+    out = tf.reshape(pred.output_tensors[0], [-1, 28*28])
+    out = tf.identity(out, 'X_out')
+    decoder = tf.graph_util.convert_variables_to_constants(pred.sess, pred.graph.as_graph_def(), ['X_out'])
+    with tf.gfile.GFile('InfoGAN-mnist.pb', "wb") as f:
+        f.write(decoder.SerializeToString())
+    return None
+
     while True:
         # only categorical turned on
         z_noise = np.random.uniform(-1, 1, (100, NOISE_DIM))

examples/GAN/steps_to_save

@@ -0,0 +1,9 @@
+# Train the model
+python InfoGAN-mnist.py
+
+# Save the model in an alternate format
+python InfoGAN-mnist.py --sample --load ./info_train/model-50000
+
+# Copy the model to the right place; from there, run the tensorflowjs converter
+# as normal.
+cp InfoGAN-mnist.pb /path/to/web/interface