Hi, I'm currently trying to run a 'ImageRecognitionInception' example with my pre-trained model (which is .pb file)
and here comes a problem :
Tensorflow.InvalidArgumentError: 'Can not squeeze dim[1], expected a dimension of 1, got 3
[[{{node InceptionV1/Logits/SpatialSqueeze}}]]'
and here is the code
using NumSharp;
using System.Collections.Generic;
using System.Diagnostics;
using System.Drawing;
using System.IO;
using Tensorflow;
using Tensorflow.Keras.Utils;
using static Tensorflow.Binding;
using Console = Colorful.Console;
namespace TensorFlowNET.Examples
{
/// Inception v3 is a widely-used image recognition model
/// that has been shown to attain greater than 78.1% accuracy on the ImageNet dataset.
/// The model is the culmination of many ideas developed by multiple researchers over the years.
public class ImageRecognitionInception : SciSharpExample, IExample
{
string dir = "ImageRecognitionInception";
string pbFile = "tensorflow_inception_graph.pb";
string labelFile = "imagenet_comp_graph_label_strings.txt";
List<NDArray> file_ndarrays = new List<NDArray>();
public ExampleConfig InitConfig()
=> Config = new ExampleConfig
{
Name = "Image Recognition Inception",
Enabled = true,
IsImportingGraph = false
};
public bool Run()
{
tf.compat.v1.disable_eager_execution();
PrepareData();
var graph = new Graph();
//import GraphDef from pb file
graph.Import(Path.Join(dir, pbFile));
var input_name = "input";
var output_name = "InceptionV1/Logits/Predictions/Softmax";
var input_operation = graph.OperationByName(input_name);
var output_operation = graph.OperationByName(output_name);
var labels = File.ReadAllLines(Path.Join(dir, labelFile));
var result_labels = new List<string>();
var sw = new Stopwatch();
using (var sess = tf.Session(graph))
{
foreach (var nd in file_ndarrays)
{
sw.Restart();
var results = sess.run(output_operation.outputs[0], (input_operation.outputs[0], nd)); // the error happens here
results = np.squeeze(results);
int idx = np.argmax(results);
Console.WriteLine($"{labels[idx]} {results[idx]} in {sw.ElapsedMilliseconds}ms", Color.Tan);
result_labels.Add(labels[idx]);
}
}
return result_labels.Contains("Blues");
}
private NDArray ReadTensorFromImageFile(string file_name,
int input_height = 288,
int input_width = 432,
int input_mean = 0,
int input_std = 255)
{
var graph = tf.Graph().as_default();
var file_reader = tf.io.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 });
using (var sess = tf.Session(graph))
return sess.run(normalized);
}
public override void PrepareData()
{
//DirectoryInfo di = new DirectoryInfo(dir);
//di.Delete(true);
Directory.CreateDirectory(dir);
// get model file
string url = "https://github.com/dorodorodororo/trainedmodel/raw/main/trainedmodel.zip"; //my own trained model
Web.Download(url, dir, "trainedmodel.zip");
Compress.UnZip(Path.Join(dir, "trainedmodel.zip"), dir);
// download sample picture
Directory.CreateDirectory(Path.Join(dir, "img"));
url = $"https://raw.githubusercontent.com/dorodorodororo/Csharp/main/blues00081.jpg";
Web.Download(url, Path.Join(dir, "img"), "blues00081.jpg");
// load image file
var files = Directory.GetFiles(Path.Join(dir, "img"));
for (int i = 0; i < files.Length; i++)
{
var nd = ReadTensorFromImageFile(files[i]);
file_ndarrays.Add(nd);
}
}
}
}
It says it expected a dimension of 1, but mine is 3.
So I brought it on Netron and got this flowchart. (It's too large so I just got the output node.)
I don't know how to fix this problem.
Can fix this problem by using a V3 model in training instead of V1? (I made this .pb file using V1 model and I think you used model using V3)
I'd appreciate it if you could give me any suggestions or comments.
Hi, I'm currently trying to run a 'ImageRecognitionInception' example with my pre-trained model (which is .pb file)
and here comes a problem :
Tensorflow.InvalidArgumentError: 'Can not squeeze dim[1], expected a dimension of 1, got 3 [[{{node InceptionV1/Logits/SpatialSqueeze}}]]'
and here is the code
It says it expected a dimension of 1, but mine is 3.
So I brought it on Netron and got this flowchart. (It's too large so I just got the output node.)
I don't know how to fix this problem.
Can fix this problem by using a V3 model in training instead of V1? (I made this .pb file using V1 model and I think you used model using V3)
I'd appreciate it if you could give me any suggestions or comments.
Thank you Have a nice day!