Sciwork demo

.gitignore

@@ -0,0 +1 @@
+uTensor

.vscode/settings.json

@@ -0,0 +1,3 @@
+{
+    "python.pythonPath": ".venv/bin/python"
+}

Pipfile

@@ -11,4 +11,4 @@ utensor-cgen = "*"
 jupyter = "*"
 
 [requires]
-python_version = "3.7"
+python_version = "3.6"

README.md

@@ -125,11 +125,18 @@ tflm_keras_export(
 $ mbed deploy
 $ mbed compile -m auto -t GCC_ARM -f --sterm
 ```
-Expected output:
 
-```
+Expected output:
+```bash
 Simple MNIST end-to-end uTensor cli example (device)
-Predicted label: 7
+pred label: 8, expecting: 8
+pred label: 3, expecting: 3
+pred label: 5, expecting: 5
+pred label: 5, expecting: 5
+pred label: 1, expecting: 1
+pred label: 9, expecting: 9
+pred label: 3, expecting: 3
+pred label: 1, expecting: 1
 ```
 
 ## Join Us

gen_inputs_header.py

@@ -0,0 +1,48 @@
+import tensorflow as tf
+import numpy as np
+import argparse
+
+
+def main(num_samples=5, seed=None):
+    mnist = tf.keras.datasets.mnist
+    (_, _), (x_test, y_test) = mnist.load_data()
+    x_test = x_test / 255.0
+
+    # Add a channels dimension
+    x_test = x_test[..., np.newaxis]
+    total_N = y_test.shape[0]
+    num_samples = min(num_samples, total_N)
+    np.random.seed(seed)
+    idxs = np.random.choice(range(total_N), num_samples, replace=False)
+    x_selected = x_test[idxs].reshape(num_samples, -1)
+    y_selected = y_test[idxs]
+    with open("input_image.h", "w") as fid:
+        fid.write("// clang-format off\n")
+        fid.write(
+            "const float arr_input_image[{}][{}] = {{\n".format(
+                x_selected.shape[0], x_selected.shape[1]
+            )
+        )
+        for i in range(x_selected.shape[0]):
+            arr = x_selected[i]
+            fid.write("    {{ {}".format(", ".join(map(str, arr))))
+            fid.write("},\n")
+        fid.write("};\n")
+        fid.write("const int ref_labels[{}] = {{\n".format(y_selected.shape[0]))
+        fid.write("    " + ", ".join(map(str, y_selected)) + "\n")
+        fid.write("};\n\n")
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "--num-samples",
+        dest="num_samples",
+        default=5,
+        help="the number of inpute samples [default: %(default)s]",
+        type=int,
+        metavar="INTEGER",
+    )
+    parser.add_argument("--seed", default=None, help="the random seed", type=int)
+    args = vars(parser.parse_args())
+    main(**args)

main.cpp

@@ -23,24 +23,26 @@ int argmax(const Tensor &logits) {
   return max_index;
 }
 
-static My_model model;
+static MyModel model;
 
-int main(void) {
+int main(int argc, char **argv) {
+  printf("\n");
   printf("Simple MNIST end-to-end uTensor cli example (device)\n");
 
-  // create the input/output tensor
-  Tensor input_image = new RomTensor({1, 28, 28, 1}, flt, arr_input_image);
-  Tensor logits = new RamTensor({1, 10}, flt);
+  size_t num_samples = *(&ref_labels + 1) - ref_labels;
+  for (size_t i = 0; i < num_samples; ++i) {
+    // create the input/output tensor
+    Tensor input_image = new RomTensor({1, 28, 28, 1}, flt, arr_input_image[i]);
+    Tensor logits = new RamTensor({1, 10}, flt);
 
-  model.set_inputs({{My_model::input_0, input_image}})
-      .set_outputs({{My_model::output_0, logits}})
-      .eval();
-
-  int max_index = argmax(logits);
-  input_image.free();
-  logits.free();
-
-  printf("pred label: %d\r\n", max_index);
+    model.set_inputs({{MyModel::input_0, input_image}})
+        .set_outputs({{MyModel::output_0, logits}})
+        .eval();
+    int max_index = argmax(logits);
+    input_image.free();
+    logits.free();
 
+    printf("pred label: %d, expecting: %d\r\n", max_index, ref_labels[i]);
+  }
   return 0;
 }

mnist_conv.ipynb

@@ -19,8 +19,8 @@
    "execution_count": 1,
    "metadata": {
     "ExecuteTime": {
-     "end_time": "2020-06-30T04:08:49.781015Z",
-     "start_time": "2020-06-30T04:08:23.608609Z"
+     "end_time": "2020-08-27T12:41:36.312890Z",
+     "start_time": "2020-08-27T12:41:31.842215Z"
     }
    },
    "outputs": [],
@@ -33,8 +33,8 @@
    "execution_count": 2,
    "metadata": {
     "ExecuteTime": {
-     "end_time": "2020-06-30T04:08:49.810814Z",
-     "start_time": "2020-06-30T04:08:49.783918Z"
+     "end_time": "2020-08-27T12:41:37.748636Z",
+     "start_time": "2020-08-27T12:41:37.713586Z"
     },
     "tags": []
    },
@@ -72,7 +72,12 @@
   {
    "cell_type": "code",
    "execution_count": 3,
-   "metadata": {},
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2020-08-27T12:41:42.661682Z",
+     "start_time": "2020-08-27T12:41:42.146260Z"
+    }
+   },
    "outputs": [],
    "source": [
     "mnist = tf.keras.datasets.mnist\n",