Corrections tutorials

Author: SANCHES-Pedro

tutorials/generative/2d_diffusion_autoencoder/2d_diffusion_autoencoder_tutorial.ipynb

@@ -107,7 +107,6 @@
     "import os\n",
     "import tempfile\n",
     "import time\n",
-    "import os\n",
     "import matplotlib.pyplot as plt\n",
     "import numpy as np\n",
     "import torch\n",
@@ -119,8 +118,6 @@
     "from monai.config import print_config\n",
     "from monai.data import DataLoader\n",
     "from monai.utils import set_determinism\n",
-    "from torch.cuda.amp import GradScaler, autocast\n",
-    "from tqdm import tqdm\n",
     "from sklearn.linear_model import LogisticRegression\n",
     "\n",
     "from generative.inferers import DiffusionInferer\n",
@@ -152,8 +149,7 @@
    "outputs": [],
    "source": [
     "directory = os.environ.get(\"MONAI_DATA_DIRECTORY\")\n",
-    "root_dir = tempfile.mkdtemp() if directory is None else directory\n",
-    "root_dir = '/home/s2086085/pedro_idcom/experiment_data'"
+    "root_dir = tempfile.mkdtemp() if directory is None else directory"
    ]
   },
   {
@@ -192,6 +188,7 @@
    ]
   },
   {
+   "attachments": {},
    "cell_type": "markdown",
    "id": "6986f55c",
    "metadata": {},
@@ -200,27 +197,19 @@
     "\n",
     "1. `LoadImaged` loads the brain images from files.\n",
     "2. `EnsureChannelFirstd` ensures the original data to construct \"channel first\" shape.\n",
-    "3.  The first `Lambdad` transform chooses the first channel of the image, which is the T1-weighted image.\n",
-    "4. `Spacingd` resamples the image to the specified voxel spacing, we use 3,3,2 mm to match the original paper.\n",
+    "3.  The first `Lambdad` transform chooses the first channel of the image, which is the Flair image.\n",
+    "4. `Spacingd` resamples the image to the specified voxel spacing, we use 3,3,2 mm.\n",
     "5. `ScaleIntensityRangePercentilesd` Apply range scaling to a numpy array based on the intensity distribution of the input. Transform is very common with MRI images.\n",
     "6. `RandSpatialCropd` randomly crop out a 2D patch from the 3D image.\n",
     "6. The last `Lambdad` transform obtains `slice_label` by summing up the label to have a single scalar value (healthy `=1` or not `=2` )."
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 5,
+   "execution_count": null,
    "id": "c68d2d91-9a0b-4ac1-ae49-f4a64edbd82a",
    "metadata": {},
-   "outputs": [
-    {
-     "name": "stderr",
-     "output_type": "stream",
-     "text": [
-      "<class 'monai.transforms.utility.array.AddChannel'>: Class `AddChannel` has been deprecated since version 0.8. please use MetaTensor data type and monai.transforms.EnsureChannelFirst instead.\n"
-     ]
-    }
-   ],
+   "outputs": [],
    "source": [
     "channel = 0  # 0 = Flair\n",
     "assert channel in [0, 1, 2, 3], \"Choose a valid channel\"\n",
@@ -229,8 +218,7 @@
     "    [\n",
     "        transforms.LoadImaged(keys=[\"image\", \"label\"]),\n",
     "        transforms.EnsureChannelFirstd(keys=[\"image\", \"label\"]),\n",
-    "        transforms.Lambdad(keys=[\"image\"], func=lambda x: x[channel, :, :, :]),\n",
-    "        transforms.AddChanneld(keys=[\"image\"]),\n",
+    "        transforms.Lambdad(keys=[\"image\"], func=lambda x: x[channel, None, :, :, :]),\n",
     "        transforms.EnsureTyped(keys=[\"image\", \"label\"]),\n",
     "        transforms.Orientationd(keys=[\"image\", \"label\"], axcodes=\"RAS\"),\n",
     "        transforms.Spacingd(keys=[\"image\", \"label\"], pixdim=(3.0, 3.0, 2.0), mode=(\"bilinear\", \"nearest\")),\n",
@@ -292,7 +280,7 @@
     "    section=\"training\",\n",
     "    cache_rate=1.0,  # you may need a few Gb of RAM... Set to 0 otherwise\n",
     "    num_workers=4,\n",
-    "    download=False,  # Set download to True if the dataset hasnt been downloaded yet\n",
+    "    download=True,  # Set download to True if the dataset hasnt been downloaded yet\n",
     "    seed=0,\n",
     "    transform=train_transforms,\n",
     ")\n",
@@ -336,7 +324,7 @@
     "    section=\"validation\",\n",
     "    cache_rate=1,  # you may need a few Gb of RAM... Set to 0 otherwise\n",
     "    num_workers=4,\n",
-    "    download=False,  # Set download to True if the dataset hasnt been downloaded yet\n",
+    "    download=True,  # Set download to True if the dataset hasnt been downloaded yet\n",
     "    seed=0,\n",
     "    transform=train_transforms,\n",
     ")\n",
@@ -645,7 +633,7 @@
     }
    ],
    "source": [
-    "n_iterations = 1e4 # training for longer helps a lot with reconstruction quality, even if the loss is already low\n",
+    "n_iterations = 1e4 # training for longer (1e4 ~ 3h) helps a lot with reconstruction quality, even if the loss is already low\n",
     "batch_size = 64\n",
     "val_interval = 100\n",
     "iter_loss_list, val_iter_loss_list = [], []\n",
@@ -810,7 +798,9 @@
     "# get latent space of training set\n",
     "latents_train = []\n",
     "classes_train = []\n",
-    "for i in range(15): # 15 slices from each volume\n",
+    "# 15 slices from each volume\n",
+    "nb_slices_per_volume = 15\n",
+    "for _ in range(nb_slices_per_volume): \n",
     "    for batch in train_loader:\n",
     "        images = batch[\"image\"].to(device)\n",
     "        latent = model.semantic_encoder(images)\n",