Image Segmentation with TensorFlow

Question

I am trying to see the possibility of using TensorFlow to identify functions in my image data. I have 50x50 pixels grayscale images that I would like to segment: the desired result will be either 0 or 1 for each pixel. 0 for the background, 1 as the core.

Input Example: Raw Input

Label example (what will be the "label" / real answer): output (label)

Can I use TensorFlow to perform this type of machine learning in my dataset? I could have thousands of images for a set of workouts.

In many examples, the label corresponds to one category, for example, to a 10-digit array [0,0,0,0,0,0,0,0,0,0,0,0,0] for a handwritten bit data set, but I have not seen many examples that would produce a large array. I would suggest that the shortcut will be a 50x50 array?

Also, any ideas on CPU processor time for this analysis time?

Answer 1

Yes , this is possible with TensorFlow. In fact, there are many ways to approach it. Here is very simple:

. . , . ( ) ( ). .

(, NN) . , .

---

, , , TensorFlow. TensorFlow, backprop, , TF .

Answer 2

SoftmaxWithLoss() , [, , , ] [N, ].

= 2, , N = batchheightwidth, SoftmaxWithLoss() .

. , .

Answer 3

. . , . softmaxWithLoss. .