Class FlowNet2S

examples/OpticalFlow/flownet_models.py:323–375

class FlowNet2S(FlowNetBase):
    def graph_structure(self, x, standalone=True):
        """
        Architecture of FlowNetSimple in Figure 2 of FlowNet 1.0.

        Args:
            x: 2CHW if standalone==True, else NCHW where C=12 is a concatenation
                of 5 tensors of [3, 3, 3, 2, 1] channels.
            standalone: If True, this model is used to predict flow from two inputs.
                If False, this model is used as part of the FlowNet2.
        """
        if standalone:
            x = tf.concat(tf.split(x, 2, axis=0), axis=1)

        with argscope([tf.layers.conv2d], activation=lambda x: tf.nn.leaky_relu(x, 0.1),
                      padding='valid', strides=2, kernel_size=3,
                      data_format='channels_first'), \
            argscope([tf.layers.conv2d_transpose], padding='same', activation=tf.identity,
                     data_format='channels_first', strides=2, kernel_size=4):
            x = tf.layers.conv2d(pad(x, 3), 64, kernel_size=7, name='conv1')
            conv2 = tf.layers.conv2d(pad(x, 2), 128, kernel_size=5, name='conv2')
            x = tf.layers.conv2d(pad(conv2, 2), 256, kernel_size=5, name='conv3')
            conv3 = tf.layers.conv2d(pad(x, 1), 256, name='conv3_1', strides=1)
            x = tf.layers.conv2d(pad(conv3, 1), 512, name='conv4')
            conv4 = tf.layers.conv2d(pad(x, 1), 512, name='conv4_1', strides=1)
            x = tf.layers.conv2d(pad(conv4, 1), 512, name='conv5')
            conv5 = tf.layers.conv2d(pad(x, 1), 512, name='conv5_1', strides=1)
            x = tf.layers.conv2d(pad(conv5, 1), 1024, name='conv6')
            conv6 = tf.layers.conv2d(pad(x, 1), 1024, name='conv6_1', strides=1)

            flow6 = tf.layers.conv2d(pad(conv6, 1), 2, name='predict_flow6', strides=1, activation=tf.identity)
            flow6_up = tf.layers.conv2d_transpose(flow6, 2, name='upsampled_flow6_to_5', use_bias=False)
            x = tf.layers.conv2d_transpose(conv6, 512, name='deconv5', activation=lambda x: tf.nn.leaky_relu(x, 0.1))

            concat5 = tf.concat([conv5, x, flow6_up], axis=1, name='concat5')
            flow5 = tf.layers.conv2d(pad(concat5, 1), 2, name='predict_flow5', strides=1, activation=tf.identity)
            flow5_up = tf.layers.conv2d_transpose(flow5, 2, name='upsampled_flow5_to_4', use_bias=False)
            x = tf.layers.conv2d_transpose(concat5, 256, name='deconv4', activation=lambda x: tf.nn.leaky_relu(x, 0.1))

            concat4 = tf.concat([conv4, x, flow5_up], axis=1, name='concat4')
            flow4 = tf.layers.conv2d(pad(concat4, 1), 2, name='predict_flow4', strides=1, activation=tf.identity)
            flow4_up = tf.layers.conv2d_transpose(flow4, 2, name='upsampled_flow4_to_3', use_bias=False)
            x = tf.layers.conv2d_transpose(concat4, 128, name='deconv3', activation=lambda x: tf.nn.leaky_relu(x, 0.1))

            concat3 = tf.concat([conv3, x, flow4_up], axis=1, name='concat3')
            flow3 = tf.layers.conv2d(pad(concat3, 1), 2, name='predict_flow3', strides=1, activation=tf.identity)
            flow3_up = tf.layers.conv2d_transpose(flow3, 2, name='upsampled_flow3_to_2', use_bias=False)
            x = tf.layers.conv2d_transpose(concat3, 64, name='deconv2', activation=lambda x: tf.nn.leaky_relu(x, 0.1))

            concat2 = tf.concat([conv2, x, flow3_up], axis=1, name='concat2')
            flow2 = tf.layers.conv2d(pad(concat2, 1), 2, name='predict_flow2', strides=1, activation=tf.identity)

            return tf.identity(flow2, name='flow2')


class FlowNet2C(FlowNetBase):