Computer Vision News - August 2016

10 Computer Vision News Tool Now, let’s see the TensorFlow version of this code 1 . Our graph will be very simple at this stage: only one node and one edge. The edge of the graph will hold the image to be processed and the node (the operation) will be our getHoughLines function. The above code declares a variable placeholder (tensor) that will hold the input image; we named it sudoku . Next, it defines a TensorFlow wrapper function (getHoughlines) as a node. Then, it reads the image into the raw_image_data variable and executes the TensorFlow session. Running this code will generate exactly the same output. So you might ask: what's the point of all this? The answer is that we are in the TensorFlow architecture and there are many benefits which we can gain from it. Amongst them: monitoring the process using the TensorBoard, easily parallelizing the process using threads, deploying computation in a parallel fashion to several CPUs, GPUs or even a cluster of computers and processing a batch of images with various initializations as parameters. How would you do all that? For this you will have to wait for Part 2 of this review of TensorFlow, which we shall publish in the September issue of Computer Vision News. There, we will continue developing and elaborating this example. We will show you also how to gain many advantages from the TensorFlow architecture beyond “deep learning” (which is, without any doubt, absolutely amazing). 1 We won’t get into the basic details of TensorFlow, since many tutorials are already available on the web for this purpose, here and here . Tool