https://github.com/tensorflow/tensorflow/issues/3332
caffe和pytorch都有对应的native API,但是TF貌似并没有,只有depth wise conv
所以现在用TF都是tf.split
和tf.concat
https://github.com/tensorflow/tensorflow/issues/3332#issuecomment-330124687
所以deploy肯定速度不是很好
https://github.com/tensorflow/tensorflow/issues/3332
caffe和pytorch都有对应的native API,但是TF貌似并没有,只有depth wise conv
所以现在用TF都是tf.split
和tf.concat
https://github.com/tensorflow/tensorflow/issues/3332#issuecomment-330124687
所以deploy肯定速度不是很好