"Machine learning" tensorflow use cases on the IOS side

Support original, more content Welcome to the author blog:
http://www.china10s.com/blog/?p=490

Machine learning This method of calculation has been known to the world in the last century, but it has not been developed because of the computer-limited computing power and network speed. With the Moore effect, the current computer performance has soared, even in the hands of the iphone, than the United States on the moon on the machine used to be stronger. Therefore, in this context, machine learning began to develop rapidly, the major companies have invested resources in this area, hoping to get a piece.

Google has always been a technology-oriented company, and now they have open-source a set of computer engine –tensorflow. He supports PCs and mobile versions, and has plenty of learning resources. It has the characteristics of high flexibility, portability, automatic differentiation, multi-language support, optimal performance and so on. It can be said to shorten the distance between the scientific research and products, eliminating a lot of repetitive code writing time.

Tensorflow™ is an open source software library that uses a data flow graph (stream graphs) for numerical computations. A node (Nodes) represents a mathematical operation in a graph, and a line (edges) in a graph represents an array of multidimensional data, the tensor (tensor), that is interconnected between nodes. Its flexible architecture allows you to expand computing on a variety of platforms, such as one or more CPUs (or GPU), servers, mobile devices, and so on in a desktop computer. TensorFlow was originally developed by researchers and engineers from the Google Brain group (Google Machine Intelligence Research) for machine learning and deep neural networks, but the versatility of the system makes it widely available in other computing areas.

Now, we can quickly deploy TensorFlow to mobile platforms, including IOS and Android platforms. Take IOS for example. Deployment: First, download tensorflow files

Since Tensoreflow has been open source in Github, it can be downloaded directly:

Github Home Address

V1.1.0 Download The second step, download the Model file

With tools, there are training models, and there is no need for us to train (and sample data).

Download the ready-made training model:

Inception v1

After the download is complete, create a new data folder in the camera Engineering directory. Copy the download to the following file:
Imagenet_comp_graph_label_strings.txt
TENSORFLOW_INCEPTION_GRAPH.PB Third step, related tools

In the absence of Libtool? , there is no successful compilation, and the installation is as follows:

sudo apt-get install Libtool

step Fourth, compile the library file

Execute the script in the following directory:

sudo./tensorflow-master/tensorflow/contrib/makefile/build_all_ios.sh

The compilation process takes about 1 hours (and machine performance).

After the compilation is complete, generate libtensorflow-core.aand copy the resulting results to the camera engineering directory
step Fifth, identify the results as shown in the figure

Cups

IPhone

Mouse

Notebook