added one more result

README.md

@@ -116,12 +116,22 @@ I'm simply using MS Paint for that. And per image, it hardly takes 1 minute to d
 
 Here are some sample overlays...
 
+![Overlay example](outputs/watermark-unavailable/overlays/overlay0.png)
+
 ![Overlay example](outputs/watermark-unavailable/overlays/overlay1.png)
 
 ![Overlay example](outputs/watermark-unavailable/overlays/overlay2.png)
 
 ## # Time for the results
 
+A sample progress of the generator:
+
+![Progress](outputs/watermark-unavailable/progress.gif)
+
+### Experiment 0
+
+![Experiment result](outputs/watermark-unavailable/output0.png)
+
 ### Experiment 1
 
 ![Experiment result](outputs/watermark-unavailable/output1.png)

notebook.ipynb

@@ -42,9 +42,8 @@
       },
       "source": [
         "def pil_to_np_array(pil_image):\n",
-        "    ar = np.array(pil_image)\n",
-        "    ar = ar.transpose(2,0,1)\n",
-        "    return ar.astype(np.float32) / 255.\n",
+        "    array = np.array(pil_image).transpose(2,0,1)\n",
+        "    return array.astype(np.float32) / 255.\n",
         "\n",
         "def np_to_torch_array(np_array):\n",
         "    return torch.from_numpy(np_array)[None, :]\n",
@@ -58,7 +57,7 @@
         "\n",
         "def save_image(np_array, step):\n",
         "    pil_image = Image.fromarray((np_array * 255.0).transpose(1, 2, 0).astype('uint8'), 'RGB')\n",
-        "    pil_image.save(f'progress-with-watermark/{str(step).zfill(len(str(TRAINING_STEPS)))}.png')\n",
+        "    pil_image.save(f'progress/{str(step).zfill(len(str(TRAINING_STEPS)))}.png')\n",
         "\n",
         "def get_image_grid(images, nrow = 3):\n",
         "    torch_images = [torch.from_numpy(x) for x in images]\n",
@@ -248,7 +247,7 @@
         "    def forward(self, x):\n",
         "        return self.model(x)\n",
         "\n",
-        "def input_noise(input_depth, spatial_size, scale = 1./10):\n",
+        "def input_noise(input_depth, spatial_size, scale = 1.0/10):\n",
         "    shape = [1, input_depth, spatial_size[0], spatial_size[1]]\n",
         "    return torch.rand(*shape) * scale"
       ],
@@ -294,9 +293,9 @@
         "tags": []
       },
       "source": [
-        "if os.path.isdir('progress-with-watermark'):\n",
-        "    rmtree('progress-with-watermark')\n",
-        "os.mkdir('progress-with-watermark')\n",
+        "if os.path.isdir('progress'):\n",
+        "    rmtree('progress')\n",
+        "os.mkdir('progress')\n",
         "\n",
         "for step in range(TRAINING_STEPS):\n",
         "    optimizer.zero_grad()\n",