Add onnx->caffe2 tutorial

bddppq · bddppq · commit 965d125af1d2 · 2017-11-17T01:55:26.000-08:00
diff --git a/tutorials/OnnxCaffe2Import.ipynb b/tutorials/OnnxCaffe2Import.ipynb
@@ -6,27 +6,89 @@
    "source": [
     "# Importing models from ONNX to Caffe2\n",
     "\n",
-    "Coming soon!"
+    "In this tutorial we are going to show you how to import ONNX models to Caffe2. You can either\n",
+    "\n",
+    "- Directly run an ONNX model with Caffe2 in Python\n",
+    "\n",
+    "or\n",
+    "\n",
+    "- Convert an ONNX model file to a Caffe2 model file, and then later run the converted Caffe2 model in any environment that Caffe2 supports, e.g. on mobile phones."
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### Installation\n",
+    "\n",
+    "The only thing you need for importing ONNX models to Caffe2 is the python package `onnx-caffe2`:\n",
+    "\n",
+    "```shell\n",
+    "$ pip install onnx-caffe2\n",
+    "```"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### Run an ONNX model with Caffe2"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 0,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import onnx\n",
+    "import onnx_caffe2.backend\n",
+    "\n",
+    "# Prepare the inputs, here we use numpy to generate some random inputs for demo purpose\n",
+    "import numpy as np\n",
+    "img = np.random.randn(1, 3, 224, 224).astype(np.float32)\n",
+    "\n",
+    "# Load the ONNX model\n",
+    "model = onnx.load('assets/squeezenet.onnx')\n",
+    "# Run the ONNX model with Caffe2\n",
+    "outputs = onnx_caffe2.backend.run_model(model, [img])"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### Convert ONNX model file to Caffe2 model file\n",
+    "`onnx-caffe2` has bundled a shell command `convert-onnx-to-caffe2` for converting ONNX model file to Caffe2 model file. \n",
+    "\n",
+    "\n",
+    "```shell\n",
+    "\n",
+    "$ convert-onnx-to-caffe2 assets/squeezenet.onnx --output predict_net.pb --init-net-output init_net.pb\n",
+    "\n",
+    "```\n",
+    "\n",
+    "Note in ONNX model file, parameters and network structure are all stored in one model file, while in Caffe2, they are normally stored in separated `init_net.pb` (parameters) and `predict_net.pb` (network structure) files."
    ]
   }
  ],
  "metadata": {
   "kernelspec": {
-   "display_name": "Python 3",
+   "display_name": "Python 2",
    "language": "python",
-   "name": "python3"
+   "name": "python2"
   },
   "language_info": {
    "codemirror_mode": {
     "name": "ipython",
-    "version": 3
+    "version": 2
    },
    "file_extension": ".py",
    "mimetype": "text/x-python",
    "name": "python",
    "nbconvert_exporter": "python",
-   "pygments_lexer": "ipython3",
-   "version": "3.6.3"
+   "pygments_lexer": "ipython2",
+   "version": "2.7.12"
   }
  },
  "nbformat": 4,
