Create app.py

app.py

@@ -0,0 +1,45 @@
+import numpy as np
+
+from PIL import Image
+from keras.models import Model
+from keras.layers import Input, UpSampling2D, Conv2D, concatenate
+
+# This is the vq-vae model from "Neural Discrete Representation Learning"
+# https://arxiv.org/abs/1711.00937
+# by Aäron van den Oord, Oriol Vinyals, Koray Kavukcuoglu (Google DeepMind)
+# ported to keras by @Ophirblum
+
+
+class Encoder:
+    def __init__(self, input_shape, latent_dim, num_embeddings, commitment_cost):
+
+        self.input_shape = input_shape
+        self.latent_dim = latent_dim
+        self.num_embeddings = num_embeddings
+        self.commitment_cost = commitment_cost
+
+        self.encoder = None
+
+    def build(self):
+
+        x = Input(shape=self.input_shape, name='encoder_input')
+
+        # Downsampling path
+
+        h = Conv2D(64, 4, strides=2, activation='relu', padding='same')(x)
+        h = Conv2D(128, 4, strides=2, activation='relu', padding='same')(h)
+        h = Conv2D(256, 4, strides=2, activation='relu', padding='same')(h)
+
+        # Latent space
+
+        z = Conv2D(self.latent_dim, 4, strides=1, activation='linear', padding='same')(h)
+
+        # Instantiate Encoder Model
+
+        self.encoder = Model(x, z)
+
+    def encode(self, x):
+
+        assert self.encoder != None, "build the encoder first"
+
+        return self.encoder.predict(x)