Created using Colaboratory

notebooks/differentiable-parameterizations/appendix/infinite_patterns.ipynb

@@ -87,11 +87,11 @@
         "id": "xy0Fs7UZ21ak",
         "colab_type": "code",
         "cellView": "form",
+        "outputId": "b09e7e58-c135-4ec7-bb4f-d5842060e473",
         "colab": {
           "base_uri": "https://localhost:8080/",
           "height": 221
-        },
-        "outputId": "b09e7e58-c135-4ec7-bb4f-d5842060e473"
+        }
       },
       "source": [
         "#@title ← Click this button to upload a different image\n",
@@ -271,7 +271,7 @@
         "\n",
         "print('\\rimage uploaded, size:', im.size, end='')"
       ],
-      "execution_count": 10,
+      "execution_count": 0,
       "outputs": [
         {
           "output_type": "display_data",
@@ -312,16 +312,19 @@
         "id": "jbPlSLt_FWwG",
         "colab_type": "code",
         "cellView": "form",
+        "outputId": "0067d7f3-a402-4240-81f0-575f187d8ba5",
         "colab": {
           "base_uri": "https://localhost:8080/",
           "height": 1041
-        },
-        "outputId": "0067d7f3-a402-4240-81f0-575f187d8ba5"
+        }
       },
       "source": [
         "#@title ← Click this button to render\n",
         "#@markdown # CPPN pattern tool\n",
-        "#@markdown  \n",
+        "#@markdown This tool uses simple neural networks that map pixel coordinates to colors to represent image, that gets optimized. This approach is also known as [Compositional pattern-producing network](https://en.wikipedia.org/wiki/Compositional_pattern-producing_network).\n",
+        "#@markdown Optimization tries to generate an image, that produces activation pattern, similar to the target image, in a particular layer (defined by **layer_index**) of ImageNet-trained classification network. **v1** objective tries to match the average pattern, while **v2** tries to match the whole distrubution.\n",
+        "#@markdown **style_weight** parameter defines the contribution of [Gatys et al](https://arxiv.org/abs/1508.06576) style loss, and ignored in case of **v2** objective.\n",
+        "#@markdown **activation** function used in the image-generating CPPN influence the resulting image style.\n",
         "\n",
         "objective = 'v2' #@param ['v1', 'v2']\n",
         "activation = 'composite' #@param ['composite', 'relu']\n",
@@ -389,7 +392,7 @@
         "img = sess.run(t_image, {t_size:1024})[0]\n",
         "show_tiled_image(img)\n"
       ],
-      "execution_count": 11,
+      "execution_count": 0,
       "outputs": [
         {
           "output_type": "display_data",
@@ -437,16 +440,16 @@
         "id": "AZ9jgnMCRBB-",
         "colab_type": "code",
         "cellView": "form",
+        "outputId": "93f9a36a-090d-46b2-b3ad-d70d800b498f",
         "colab": {
           "base_uri": "https://localhost:8080/",
           "height": 989
-        },
-        "outputId": "93f9a36a-090d-46b2-b3ad-d70d800b498f"
+        }
       },
       "source": [
         "#@title ← Click this button to render\n",
         "#@markdown # DeepDream pattern tool\n",
-        "#@markdown  \n",
+        "#@markdown This tool uses pixel based image representation and a [DeepDream-inspired](https://github.com/tensorflow/tensorflow/tree/master/tensorflow/examples/tutorials/deepdream) multiscale pattern generation approach.\n",
         "\n",
         "preview_octave_n = 6\n",
         "preview_octave_steps = 50\n",
@@ -619,7 +622,7 @@
         "  clear_output()\n",
         "  show_tiled_image(dream.rgb)"
       ],
-      "execution_count": 16,
+      "execution_count": 0,
       "outputs": [
         {
           "output_type": "display_data",