In my previous blog, I showed how we can use OpenCV and numpy to detect faces with Haar Cascade library which made us the CPU usage of 280%. Today I am gonna explain to use TensorFlow’s object detection example to detect faces. Before actually heading to the implementation, we need to know basic TF commands to deal with the actual process involved.
As We know the TensorFlow deals lot with Tensors, which are multi-dimensional arrays which hold the numerical data over computation. When TensorFlow statements executed it actually generates Computational Graph in memory. TF can be run on either CPU or GPU, both are supported. We can execute some part of the graph to run on which hardware and be back to come to the original state to complete the whole process.
These are the statements that we must know to get started:
(i) tf.Graph(), tf.GraphDef(), Gfile, tf.Session(), get_tensor_by_name(), visualize_boxes_and_labels_on_image_array
MLNOTES 