Upload image_dataset.ipynb

image_dataset.ipynb

@@ -581,7 +581,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 19,
+   "execution_count": 22,
    "metadata": {},
    "outputs": [
     {
@@ -589,14 +589,8 @@
      "output_type": "stream",
      "text": [
       "Reading metadata from /fsx/avijit/projects/datacommonsMA/massgis_2023_aerial_imagery/imagefolder-dataset/train/metadata.csv\n",
-      "Selected 50 samples\n"
-     ]
-    },
-    {
-     "name": "stderr",
-     "output_type": "stream",
-     "text": [
-      "Processing images:   0%|                                                  | 0/50 [00:00<?, ?it/s]"
+      "Selected 50 samples\n",
+      "Loading and resizing images...\n"
      ]
     },
     {
@@ -604,28 +598,20 @@
      "output_type": "stream",
      "text": [
       "/fsx/avijit/anaconda3/envs/py312/lib/python3.12/site-packages/PIL/Image.py:3402: DecompressionBombWarning: Image size (100000000 pixels) exceeds limit of 89478485 pixels, could be decompression bomb DOS attack.\n",
-      "  warnings.warn(\n",
-      "Processing images: 100%|█████████████████████████████████████████| 50/50 [00:48<00:00,  1.04it/s]"
+      "  warnings.warn(\n"
      ]
     },
     {
      "name": "stdout",
      "output_type": "stream",
      "text": [
+      "Resizing images...\n",
+      "Saving to /fsx/avijit/projects/datacommonsMA/massgis_2023_aerial_imagery/imagefolder-dataset/data/sample_dataset_256x256.parquet...\n",
       "Saved sample dataset to /fsx/avijit/projects/datacommonsMA/massgis_2023_aerial_imagery/imagefolder-dataset/data/sample_dataset_256x256.parquet\n",
-      "Processed images saved in /fsx/avijit/projects/datacommonsMA/massgis_2023_aerial_imagery/imagefolder-dataset/data/processed_images\n",
-      "File size: 0.00 MB\n",
-      "Schema of the created Parquet file:\n",
-      "image: string\n",
-      "tilename: string\n",
-      "zone: int64\n"
-     ]
-    },
-    {
-     "name": "stderr",
-     "output_type": "stream",
-     "text": [
-      "\n"
+      "File size: 6.28 MB\n",
+      "Verifying saved file...\n",
+      "Columns in saved file: ['image', 'tilename', 'zone']\n",
+      "Number of rows: 50\n"
      ]
     }
    ],
@@ -637,10 +623,17 @@
     "from PIL import Image as PILImage\n",
     "import numpy as np\n",
     "from tqdm import tqdm\n",
+    "import io\n",
+    "\n",
+    "# Import the pandas_image_methods library\n",
+    "from pandas_image_methods import PILMethods\n",
+    "\n",
+    "# Register the PIL methods accessor\n",
+    "pd.api.extensions.register_series_accessor(\"pil\")(PILMethods)\n",
     "\n",
     "base_dir = \"/fsx/avijit/projects/datacommonsMA/massgis_2023_aerial_imagery/imagefolder-dataset\"\n",
     "train_dir = os.path.join(base_dir, \"train\")\n",
-    "output_dir = os.path.join(base_dir, \"data\")  # Matches the preview config\n",
+    "output_dir = os.path.join(base_dir, \"data\")\n",
     "output_path = os.path.join(output_dir, \"sample_dataset_256x256.parquet\")\n",
     "\n",
     "# Create the output directory if it doesn't exist\n",
@@ -659,84 +652,54 @@
     "\n",
     "print(f\"Selected {len(metadata_df)} samples\")\n",
     "\n",
-    "# Initialize lists to store data\n",
-    "image_paths = []\n",
-    "tilenames = []\n",
-    "zones = []\n",
-    "\n",
-    "processed_images_dir = os.path.join(output_dir, \"processed_images\")\n",
-    "os.makedirs(processed_images_dir, exist_ok=True)\n",
-    "\n",
-    "# Process each image\n",
-    "for _, row in tqdm(metadata_df.iterrows(), total=len(metadata_df), desc=\"Processing images\"):\n",
-    "    try:\n",
-    "        # Get image path\n",
-    "        img_path = os.path.join(train_dir, row['file_name'])\n",
-    "        \n",
-    "        # Open, resize, and save the processed image\n",
-    "        with PILImage.open(img_path) as img:\n",
-    "            # Convert to RGB if needed\n",
-    "            if img.mode != 'RGB':\n",
-    "                img = img.convert('RGB')\n",
-    "            \n",
-    "            # Resize to target size\n",
-    "            img_resized = img.resize(target_size)\n",
-    "            \n",
-    "            # Save the resized image in a separate folder\n",
-    "            output_image_path = os.path.join(processed_images_dir, row['file_name'])\n",
-    "            img_resized.save(output_image_path, format='JPEG', quality=90)\n",
-    "            \n",
-    "            # Append data\n",
-    "            image_paths.append(output_image_path)  # Store file path instead of bytes\n",
-    "            tilenames.append(row['tilename'])\n",
-    "            zones.append(int(row['zone']))  # Ensure zone is an integer\n",
-    "    except Exception as e:\n",
-    "        print(f\"Error processing {row['file_name']}: {e}\")\n",
+    "# Create DataFrame with just the paths first\n",
+    "df = pd.DataFrame({\n",
+    "    'file_path': [os.path.join(train_dir, row['file_name']) for _, row in metadata_df.iterrows()],\n",
+    "    'tilename': metadata_df['tilename'].tolist(),\n",
+    "    'zone': metadata_df['zone'].astype('int64').tolist()\n",
+    "})\n",
     "\n",
-    "# Convert lists to Arrow arrays\n",
-    "image_array = pa.array(image_paths, type=pa.string())  # Store paths as strings\n",
-    "tilename_array = pa.array(tilenames, type=pa.string())\n",
-    "zone_array = pa.array(zones, type=pa.int64())\n",
+    "# Load images using the pil accessor\n",
+    "print(\"Loading and resizing images...\")\n",
+    "df['image'] = df['file_path'].pil.open()\n",
     "\n",
-    "# Define schema explicitly\n",
-    "schema = pa.schema([\n",
-    "    ('image', pa.string()),  # Store as file path (datasets library will auto-load)\n",
-    "    ('tilename', pa.string()),\n",
-    "    ('zone', pa.int64())\n",
-    "])\n",
+    "# Resize the images\n",
+    "print(\"Resizing images...\")\n",
+    "df['image'] = df['image'].pil.resize(target_size)\n",
     "\n",
-    "# Create Arrow Table using the schema\n",
-    "table = pa.Table.from_arrays([image_array, tilename_array, zone_array], schema=schema)\n",
+    "# Keep only the required columns for the preview\n",
+    "df = df[['image', 'tilename', 'zone']]\n",
     "\n",
-    "# Write to Parquet\n",
-    "pq.write_table(table, output_path, compression='snappy', flavor=['spark'])\n",
+    "# Save to Parquet (the library will handle the PIL images correctly)\n",
+    "print(f\"Saving to {output_path}...\")\n",
+    "df.to_parquet(output_path)\n",
     "\n",
     "print(f\"Saved sample dataset to {output_path}\")\n",
-    "print(f\"Processed images saved in {processed_images_dir}\")\n",
     "print(f\"File size: {os.path.getsize(output_path) / (1024 * 1024):.2f} MB\")\n",
     "\n",
-    "# To check if the schema matches what's expected\n",
-    "print(\"Schema of the created Parquet file:\")\n",
-    "parquet_schema = pq.read_schema(output_path)\n",
-    "print(parquet_schema)\n"
+    "# Verify the saved file\n",
+    "print(\"Verifying saved file...\")\n",
+    "df_check = pd.read_parquet(output_path)\n",
+    "print(\"Columns in saved file:\", df_check.columns.tolist())\n",
+    "print(\"Number of rows:\", len(df_check))"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 20,
+   "execution_count": 24,
    "metadata": {},
    "outputs": [
     {
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "<pyarrow._parquet.FileMetaData object at 0x7fa0858cc590>\n",
+      "<pyarrow._parquet.FileMetaData object at 0x7fa0fc694f90>\n",
       "  created_by: parquet-cpp-arrow version 19.0.0\n",
-      "  num_columns: 3\n",
+      "  num_columns: 4\n",
       "  num_rows: 50\n",
       "  num_row_groups: 1\n",
       "  format_version: 2.6\n",
-      "  serialized_size: 943\n"
+      "  serialized_size: 2731\n"
      ]
     }
    ],
@@ -758,6 +721,18 @@
    "display_name": "py312",
    "language": "python",
    "name": "py312"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.12.9"
   }
  },
  "nbformat": 4,