DDFlow

网络结构：

*特征提取：

```

feature_extractor: {'conv6_1': <tf.Tensor 'feature_extractor/conv6_1/Maximum:0' shape=(?, 5, 14, 192) dtype=float32>, 'conv5_1': <tf.Tensor 'feature_extractor/conv5_1/Maximum:0' shape=(?, 10, 28, 128) dtype=float32>, 'conv4_1': <tf.Tensor 'feature_extractor/conv4_1/Maximum:0' shape=(?, 20, 56, 96) dtype=float32>, 'conv2_2': <tf.Tensor 'feature_extractor/conv2_2/Maximum:0' shape=(?, 80, 224, 32) dtype=float32>, 'conv3_2': <tf.Tensor 'feature_extractor/conv3_2/Maximum:0' shape=(?, 40, 112, 64) dtype=float32>, 'conv1_1': <tf.Tensor 'feature_extractor/conv1_1/Maximum:0' shape=(?, 160, 448, 16) dtype=float32>, 'conv1_2': <tf.Tensor 'feature_extractor/conv1_2/Maximum:0' shape=(?, 160, 448, 16) dtype=float32>, 'conv6_2': <tf.Tensor 'feature_extractor/conv6_2/Maximum:0' shape=(?, 5, 14, 192) dtype=float32>, 'conv3_1': <tf.Tensor 'feature_extractor/conv3_1/Maximum:0' shape=(?, 40, 112, 64) dtype=float32>, 'conv5_2': <tf.Tensor 'feature_extractor/conv5_2/Maximum:0' shape=(?, 10, 28, 128) dtype=float32>, 'conv2_1': <tf.Tensor 'feature_extractor/conv2_1/Maximum:0' shape=(?, 80, 224, 32) dtype=float32>, 'conv4_2': <tf.Tensor 'feature_extractor/conv4_2/Maximum:0' shape=(?, 20, 56, 96) dtype=float32>}

```