6dad66511b8432570eb53b67b8fd889913ac15e3,examples/opensets/mnist_model2.py,MyModel,_build,#MyModel#,17

Before Change

    def _build(self, *args, **kwargs):
        images_shape = [None] + list(self.get_from_config("images_shape"))

        input_images = tf.placeholder("uint8", images_shape, name="input_images")
        input_labels = tf.placeholder("uint8", [None], name="input_labels")
        images = tf.to_float(input_images)

        features = conv2d_block(images, 32, 3, layout="canp", name="layer1")
        features = flatten(features)

        layer1 = tf.layers.dense(features, units=512, activation=tf.nn.relu)
        model_output = tf.layers.dense(layer1, units=10)
        predictions = tf.identity(model_output, name="predictions")

        targets = tf.one_hot(input_labels, depth=10, name="targets")
        predicted_labels = tf.argmax(model_output, axis=1, name="predicted_labels")

After Change

        placeholders, inputs = self._make_inputs(names)

        num_classes = self.num_classes("labels")
        x = conv2d_block(inputs["images"], 32, 3, layout="cnap", name="layer1")
        x = conv2d_block(x, 64, 3, layout="cnap", name="layer2")
        //x = conv2d_block(x, 128, 3, layout="cnap", name="layer3")
        x = conv2d_block(x, num_classes, 3, layout="cnap", name="layer4")

Instances