Merge pull request #361 from BhammarArjun/stage

Fixing One hot encoding part

notebooks/Unit 3 - Vision Transformers/fine-tuning-multilabel-image-classification.ipynb

@@ -579,29 +579,49 @@
    "outputs": [],
    "source": [
     "def train_transforms(batch):\n",
-    "    # convert all images in batch to RGB to avoid grayscale or transparent images\n",
-    "    batch['image'] = [x.convert('RGB') for x in batch['image']]\n",
-    "    # apply torchvision.transforms per sample in the batch\n",
-    "    inputs = [train_tfms(x) for x in batch['image']]\n",
-    "    batch['pixel_values'] = inputs\n",
-    "    \n",
-    "    # one-hot encoding the labels\n",
-    "    labels = torch.tensor(batch['classes'])\n",
-    "    batch['labels'] = nn.functional.one_hot(labels,num_classes=20).sum(dim=1)\n",
-    "    \n",
+    "    # Convert all images to RGB format\n",
+    "    if isinstance(batch['image'], list):\n",
+    "        # Batch processing\n",
+    "        batch['image'] = [x.convert('RGB') for x in batch['image']]\n",
+    "        inputs = [train_tfms(x) for x in batch['image']]\n",
+    "        batch['pixel_values'] = torch.stack(inputs)  # Stack tensor outputs\n",
+    "    else:\n",
+    "        # Single sample processing\n",
+    "        batch['image'] = batch['image'].convert('RGB')\n",
+    "        batch['pixel_values'] = train_tfms(batch['image'])\n",
+    "\n",
+    "    # One-hot encode the multilabels\n",
+    "    all_labels = [torch.tensor(labels) for labels in batch['classes']]\n",
+    "\n",
+    "    # Create one-hot encoding for each image's classes\n",
+    "    one_hot_labels = [nn.functional.one_hot(label, num_classes=20).sum(dim=0) for label in all_labels]\n",
+    "\n",
+    "    # Stack them into a batch\n",
+    "    batch['labels'] = torch.stack(one_hot_labels)\n",
+    "\n",
     "    return batch\n",
     "\n",
     "def valid_transforms(batch):\n",
-    "    # convert all images in batch to RGB to avoid grayscale or transparent images\n",
-    "    batch['image'] = [x.convert('RGB') for x in batch['image']]\n",
-    "    # apply torchvision.transforms per sample in the batch\n",
-    "    inputs = [valid_tfms(x) for x in batch['image']]\n",
-    "    batch['pixel_values'] = inputs\n",
-    "    \n",
-    "    # one-hot encoding the labels\n",
-    "    labels = torch.tensor(batch['classes'])\n",
-    "    batch['labels'] = nn.functional.one_hot(labels,num_classes=20).sum(dim=1)\n",
-    "    \n",
+    "    # Convert all images to RGB format\n",
+    "    if isinstance(batch['image'], list):\n",
+    "        # Batch processing\n",
+    "        batch['image'] = [x.convert('RGB') for x in batch['image']]\n",
+    "        inputs = [train_tfms(x) for x in batch['image']]\n",
+    "        batch['pixel_values'] = torch.stack(inputs)  # Stack tensor outputs\n",
+    "    else:\n",
+    "        # Single sample processing\n",
+    "        batch['image'] = batch['image'].convert('RGB')\n",
+    "        batch['pixel_values'] = train_tfms(batch['image'])\n",
+    "\n",
+    "    # One-hot encode the multilabels\n",
+    "    all_labels = [torch.tensor(labels) for labels in batch['classes']]\n",
+    "\n",
+    "    # Create one-hot encoding for each image's classes\n",
+    "    one_hot_labels = [nn.functional.one_hot(label, num_classes=20).sum(dim=0) for label in all_labels]\n",
+    "\n",
+    "    # Stack them into a batch\n",
+    "    batch['labels'] = torch.stack(one_hot_labels)\n",
+    "\n",
     "    return batch"
    ]
   },