Rerun BSD68 example with and without tta.

Author: tibuch

examples/2D/denoising2D_BSD68/BSD68_reproducibility.ipynb

@@ -184,7 +184,7 @@
      "name": "stderr",
      "output_type": "stream",
      "text": [
-      "/home/tbuchhol/Gitrepos/n2v/n2v/models/n2v_standard.py:405: UserWarning: output path for model already exists, files may be overwritten: /home/tbuchhol/Gitrepos/n2v/examples/2D/denoising2D_BSD68/models/BSD68_reproducability_5x5\n",
+      "/home/tbuchhol/Gitrepos/n2v/n2v/models/n2v_standard.py:416: UserWarning: output path for model already exists, files may be overwritten: /home/tbuchhol/Gitrepos/n2v/examples/2D/denoising2D_BSD68/models/BSD68_reproducability_5x5\n",
       "  'output path for model already exists, files may be overwritten: %s' % str(self.logdir.resolve()))\n",
       "Using TensorFlow backend.\n"
      ]
@@ -396,13 +396,7 @@
       "400/400 [==============================] - 117s 292ms/step - loss: 0.1924 - val_loss: 0.1916\n",
       "Epoch 74/200\n",
       "400/400 [==============================] - 117s 292ms/step - loss: 0.1937 - val_loss: 0.1719\n",
-      "Epoch 75/200\n"
-     ]
-    },
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
+      "Epoch 75/200\n",
       "400/400 [==============================] - 117s 292ms/step - loss: 0.1921 - val_loss: 0.1735\n",
       "\n",
       "Epoch 00075: ReduceLROnPlateau reducing learning rate to 9.999999747378752e-05.\n",
@@ -797,7 +791,7 @@
     "pred = []\n",
     "psnrs = []\n",
     "for gt, img in zip(groundtruth_data, test_data):\n",
-    "    p_ = model.predict(img.astype(np.float32), 'YX');\n",
+    "    p_ = model.predict(img.astype(np.float32), 'YX', tta=False);\n",
     "    pred.append(p_)\n",
     "    psnrs.append(PSNR(gt, p_))\n",
     "\n",
@@ -815,17 +809,54 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "PSNR: 27.81\n"
+      "PSNR (without test-time augmentation): 27.81\n"
      ]
     }
    ],
    "source": [
-    "print(\"PSNR:\", np.round(np.mean(psnrs), 2))"
+    "print(\"PSNR (without test-time augmentation):\", np.round(np.mean(psnrs), 2))"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": 15,
+   "metadata": {
+    "scrolled": true
+   },
+   "outputs": [],
+   "source": [
+    "pred = []\n",
+    "psnrs = []\n",
+    "for gt, img in zip(groundtruth_data, test_data):\n",
+    "    p_ = model.predict(img.astype(np.float32), 'YX', tta=True);\n",
+    "    pred.append(p_)\n",
+    "    psnrs.append(PSNR(gt, p_))\n",
+    "\n",
+    "psnrs = np.array(psnrs)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 16,
+   "metadata": {
+    "scrolled": true
+   },
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "PSNR (with test-time augmentation): 27.88\n"
+     ]
+    }
+   ],
+   "source": [
+    "print(\"PSNR (with test-time augmentation):\", np.round(np.mean(psnrs), 2))"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 17,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -835,14 +866,14 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 16,
+   "execution_count": 18,
    "metadata": {},
    "outputs": [],
    "source": [
     "pred = []\n",
     "psnrs = []\n",
     "for gt, img in zip(groundtruth_data, test_data):\n",
-    "    p_ = model.predict(img.astype(np.float32), 'YX')\n",
+    "    p_ = model.predict(img.astype(np.float32), 'YX', tta=False)\n",
     "    pred.append(p_)\n",
     "    psnrs.append(PSNR(gt, p_))\n",
     "\n",
@@ -851,7 +882,44 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 17,
+   "execution_count": 19,
+   "metadata": {
+    "scrolled": true
+   },
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "PSNR (without test-time augmentation): 27.77\n"
+     ]
+    }
+   ],
+   "source": [
+    "print(\"PSNR (without test-time augmentation):\", np.round(np.mean(psnrs), 2))"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 20,
+   "metadata": {
+    "scrolled": true
+   },
+   "outputs": [],
+   "source": [
+    "pred = []\n",
+    "psnrs = []\n",
+    "for gt, img in zip(groundtruth_data, test_data):\n",
+    "    p_ = model.predict(img.astype(np.float32), 'YX', tta=True)\n",
+    "    pred.append(p_)\n",
+    "    psnrs.append(PSNR(gt, p_))\n",
+    "\n",
+    "psnrs = np.array(psnrs)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 21,
    "metadata": {
     "scrolled": true
    },
@@ -860,12 +928,12 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "PSNR: 27.77\n"
+      "PSNR (with test-time augmentation): 27.84\n"
      ]
     }
    ],
    "source": [
-    "print(\"PSNR:\", np.round(np.mean(psnrs), 2))"
+    "print(\"PSNR (with test-time augmentation):\", np.round(np.mean(psnrs), 2))"
    ]
   }
  ],