using NumSharp; using System; using System.Collections.Generic; using System.IO; using System.IO.Compression; using System.Linq; using System.Net; using System.Text; using Tensorflow; namespace TensorFlowNET.Examples { public class ImageRecognition : Python, IExample { public int Priority => 7; public bool Enabled { get; set; } = true; public string Name => "Image Recognition"; string dir = "ImageRecognition"; string pbFile = "tensorflow_inception_graph.pb"; string labelFile = "imagenet_comp_graph_label_strings.txt"; string picFile = "grace_hopper.jpg"; public bool Run() { PrepareData(); var labels = File.ReadAllLines(Path.Join(dir, labelFile)); var files = Directory.GetFiles(Path.Join(dir, "img")); foreach (var file in files) { var tensor = ReadTensorFromImageFile(file); var graph = new Graph().as_default(); //import GraphDef from pb file graph.Import(Path.Join(dir, pbFile)); var input_name = "input"; var output_name = "output"; var input_operation = graph.OperationByName(input_name); var output_operation = graph.OperationByName(output_name); var idx = 0; float propability = 0; with(tf.Session(graph), sess => { var results = sess.run(output_operation.outputs[0], new FeedItem(input_operation.outputs[0], tensor)); var probabilities = results.Data(); for (int i = 0; i < probabilities.Length; i++) { if (probabilities[i] > propability) { idx = i; propability = probabilities[i]; } } }); Console.WriteLine($"{picFile}: {labels[idx]} {propability}"); return labels[idx].Equals("military uniform"); } return false; } private NDArray ReadTensorFromImageFile(string file_name, int input_height = 224, int input_width = 224, int input_mean = 117, int input_std = 1) { return with(tf.Graph().as_default(), graph => { var file_reader = tf.read_file(file_name, "file_reader"); var decodeJpeg = tf.image.decode_jpeg(file_reader, channels: 3, name: "DecodeJpeg"); var cast = tf.cast(decodeJpeg, tf.float32); var dims_expander = tf.expand_dims(cast, 0); var resize = tf.constant(new int[] { input_height, input_width }); var bilinear = tf.image.resize_bilinear(dims_expander, resize); var sub = tf.subtract(bilinear, new float[] { input_mean }); var normalized = tf.divide(sub, new float[] { input_std }); return with(tf.Session(graph), sess => sess.run(normalized)); }); } public void PrepareData() { Directory.CreateDirectory(dir); // get model file string url = "https://storage.googleapis.com/download.tensorflow.org/models/inception5h.zip"; Utility.Web.Download(url, dir, "inception5h.zip"); Utility.Compress.UnZip(Path.Join(dir, "inception5h.zip"), dir); // download sample picture Directory.CreateDirectory(Path.Join(dir, "img")); url = $"https://raw.githubusercontent.com/tensorflow/tensorflow/master/tensorflow/examples/label_image/data/grace_hopper.jpg"; Utility.Web.Download(url, Path.Join(dir, "img"), "grace_hopper.jpg"); } } }