tfrun

commit f8cc5c2938092ba0ccbad87e4ae8bafa2b34a5aa
parent fc6ef2c6169881140bc4b7f7d92b831e9159fa61
Author: Mohammad-Reza Nabipoor <mnabipoor@gnu.org>
Date:   Tue,  3 Aug 2021 03:09:44 +0430

Add README

Diffstat:
A
README.md
 | 
46
++++++++++++++++++++++++++++++++++++++++++++++

1 file changed, 46 insertions(+), 0 deletions(-)
diff --git a/README.md b/README.md
@@ -0,0 +1,46 @@
+
+# `tfrun`
+
+An easy-to-use C++ wrapper over the stable C API of TensorFlow.
+
+## How to use?
+
+```c++
+void
+tfrun_example(const char* model_path)
+{
+  // `model_path` is the path of a pre-trained MobileNetV2 network
+  tfrun::siso net{ model_path, "input", "MobilenetV2/Predictions/Reshape_1" };
+  enum
+  {
+    ndims = 4,
+    nclasses = 1000 + 1 /*background*/,
+    nbatch = 1,
+  };
+  std::array<int64_t, ndims> dims{ nbatch, 224, 224, 3 };
+  int width;
+  int height;
+  int chan;
+  std::vector<float> img;
+
+  img = image(bmpread("dog.bmp", &width, &height, &chan));
+  assert(img.size() == nbatch * 224 * 224 * 3);
+
+  auto out =
+    net.run(dims.data(), ndims, img.data(), img.size() * sizeof(img[0]));
+  auto it = std::max_element(out.data.begin(), out.data.end());
+  auto classIdx = it - out.data.begin();
+
+  assert(out.dims == std::vector<int64_t>{
+                       nbatch,
+                       nclasses,
+                     });
+  assert(classIdx == 1 + 162 /*beagle*/);
+}
+```
+
+## How to run tests?
+
+Please download the pre-trained model and set the environment variable
+`MODEL_PATH` to the path of downloaded model and run the `test/tfrun.test`
+program.