{ "nbformat": 4, "nbformat_minor": 0, "metadata": { "colab": { "provenance": [] }, "kernelspec": { "name": "python3", "display_name": "Python 3" }, "language_info": { "name": "python" } }, "cells": [ { "cell_type": "code", "execution_count": null, "metadata": { "id": "0fsKoLM5sjn_" }, "outputs": [], "source": [ "# Colab Setup Cell\n", "# Install necessary packages\n", "!pip -q install -U datasets==3.1.0 pillow pyarrow matplotlib tqdm huggingface_hub\n", "\n", "from pathlib import Path\n", "from datasets import load_dataset, Image\n", "import matplotlib.pyplot as plt\n", "import matplotlib.patches as patches\n", "from PIL import Image as PILImage\n", "from tqdm.auto import tqdm\n", "from huggingface_hub import hf_hub_url, notebook_login, hf_hub_download # Use hf_hub_download\n", "import os # Necessary for creating folders\n", "\n", "# --- HUGGING FACE AUTHENTICATION (Keep this) ---\n", "print(\"--- Authenticating with Hugging Face ---\")\n", "notebook_login()\n", "\n", "# --- Configuration ---\n", "DATASET_ID = \"c-i-ber/Nova\"\n", "SPLIT_NAME = \"train\"\n", "PARQUET_FILE_URL = f\"hf://datasets/{DATASET_ID}/data/nova-v1.parquet\"\n", "\n", "# --- LOCAL FOLDER CONSTANT ---\n", "LOCAL_IMAGE_DIR = \"images\" # The folder where images will be saved\n", "\n", "MAX_BOXES_TO_DRAW = 5\n", "\n", "# --- Visualization Style Constants (Unchanged) ---\n", "BOX_COLOR = 'cyan'\n", "TEXT_COLOR = 'black'\n", "BOX_LABEL_BG_COLOR = 'white'\n", "TITLE_FONT_SIZE = 12\n", "METADATA_SECTION_TITLE_SIZE = 12\n", "METADATA_TEXT_SIZE = 10" ] }, { "cell_type": "code", "source": [ "# --- HUGGING FACE AUTHENTICATION FIX ---\n", "from huggingface_hub import notebook_login\n", "import os\n", "\n", "# 1. Log in to Hugging Face\n", "print(\"--- Authenticating with Hugging Face ---\")\n", "# This command will open an interactive prompt where you paste your token.\n", "# You can get a token from: https://huggingface.co/settings/tokens\n", "notebook_login()" ], "metadata": { "id": "yRAVEjw-28yU" }, "execution_count": null, "outputs": [] }, { "cell_type": "code", "source": [ "def load_and_prepare_dataset(parquet_url: str) -> \"datasets.Dataset\":\n", " \"\"\"\n", " Loads the dataset structure (metadata) directly from the remote Parquet file.\n", " \"\"\"\n", " print(f\"Loading metadata structure from: {parquet_url}\")\n", " try:\n", " ds = load_dataset(\"parquet\", data_files=parquet_url, split=SPLIT_NAME)\n", " except Exception as e:\n", " print(f\"\\nšŸšØ Fatal Error: Could not load Parquet file from URL. Details: {e}\")\n", " return None\n", "\n", " print(f\"\\nDataset loaded successfully. Total examples: {len(ds)}\")\n", " if \"bboxes\" in ds.column_names:\n", " print(\"āœ… SUCCESS: All metadata columns loaded correctly.\")\n", " else:\n", " print(\"šŸšØ CRITICAL ERROR: 'bboxes' column is missing.\")\n", "\n", " return ds\n", "\n", "\n", "def cache_all_images(dataset: \"datasets.Dataset\", dataset_id: str):\n", " \"\"\"\n", " Downloads all images from the HF Hub to the local folder and saves them as JPEGs.\n", " \"\"\"\n", " Path(LOCAL_IMAGE_DIR).mkdir(exist_ok=True)\n", " print(f\"\\n--- Downloading and Caching All Images to: {LOCAL_IMAGE_DIR}/ ---\")\n", "\n", " for example in tqdm(dataset, desc=\"Downloading and saving images\", unit=\"img\"):\n", " try:\n", " # 1. Get the image's remote path and local target path\n", " image_url_path = example['image_path']\n", "\n", " # The filename is the part after the last slash\n", " filename = Path(image_url_path).name\n", " local_save_path = Path(LOCAL_IMAGE_DIR) / filename\n", "\n", " # Skip if the file already exists locally\n", " if local_save_path.exists():\n", " continue\n", "\n", " # 2. Download the file's absolute path from HF cache\n", " # This uses the correct function and full internal path\n", " repo_file_path = image_url_path\n", " cached_file_path = hf_hub_download(\n", " repo_id=dataset_id,\n", " filename=repo_file_path,\n", " repo_type=\"dataset\"\n", " )\n", "\n", " # 3. Open the downloaded file from the cache and save it to the local folder\n", " img = PILImage.open(cached_file_path)\n", " img.save(local_save_path)\n", "\n", " except Exception as e:\n", " print(f\"\\nWarning: Could not process image {example.get('filename')}. Error: {e}\")\n", "\n", " print(\"Caching complete. Demo examples will load instantly from the local folder.\")\n", "\n", "\n", "def load_local_image(filename: str) -> PILImage.Image:\n", " \"\"\"Loads an image from the local directory using its simple filename.\"\"\"\n", " local_path = Path(LOCAL_IMAGE_DIR) / filename\n", " return PILImage.open(local_path)\n", "\n", "\n", "# --- Example Selector Functions (Unchanged) ---\n", "# ... (has_boxes, get_example_by_filename, find_first_example_with_boxes are here) ...\n", "\n", "def has_boxes(example: dict) -> bool:\n", " bboxes = example.get(\"bboxes\", None)\n", " return isinstance(bboxes, list) and len(bboxes) > 0\n", "\n", "def get_example_by_filename(dataset: \"datasets.Dataset\", filename: str) -> dict:\n", " \"\"\"Retrieves an example from the dataset by its 'filename'.\"\"\"\n", " for i, ex in tqdm(\n", " enumerate(dataset),\n", " total=len(dataset),\n", " desc=f\"Searching for '{filename}'\",\n", " unit=\"ex\",\n", " leave=False\n", " ):\n", " if ex.get(\"filename\") == filename:\n", " return dataset[i]\n", " raise ValueError(f\"Example with filename '{filename}' not found.\")\n", "\n", "def find_first_example_with_boxes(dataset: \"datasets.Dataset\") -> dict:\n", " \"\"\"Finds the first example in the dataset that contains bounding boxes.\"\"\"\n", " print(\"Searching for first example with bounding boxes...\")\n", " for ex in tqdm(dataset, desc=\"Finding example with boxes\", unit=\"ex\"):\n", " if has_boxes(ex):\n", " return ex\n", "\n", " print(\"Warning: No example with bounding boxes found. Displaying index 0 instead.\")\n", " return dataset[0]" ], "metadata": { "id": "5u2Ezk2YFh3o" }, "execution_count": null, "outputs": [] }, { "cell_type": "code", "source": [ "# Execute the loading\n", "ds = load_and_prepare_dataset(PARQUET_FILE_URL)\n", "\n", "if ds is None:\n", " raise RuntimeError(\"Dataset loading failed. Cannot proceed with demo.\")\n", "\n", "# --- STEP 2: CACHE ALL IMAGES ---\n", "cache_all_images(ds, DATASET_ID)" ], "metadata": { "id": "hPB6_pzXIBup" }, "execution_count": null, "outputs": [] }, { "cell_type": "code", "source": [ "def display_nova_example(dataset: \"datasets.Dataset\", selector: str | int = None):\n", " # --- Selection Logic (Unchanged) ---\n", " example = None\n", " try:\n", " if isinstance(selector, int):\n", " if 0 <= selector < len(dataset):\n", " example = dataset[selector]\n", " else:\n", " raise IndexError(f\"Index {selector} out of bounds for dataset of size {len(dataset)}.\")\n", " elif isinstance(selector, str):\n", " example = get_example_by_filename(dataset, selector)\n", " elif selector is None:\n", " example = find_first_example_with_boxes(dataset)\n", " else:\n", " raise TypeError(f\"Selector must be an integer index or a string filename, got {type(selector)}.\")\n", "\n", " if example is None: return\n", "\n", " filename = example.get('filename', 'N/A')\n", " total_boxes = len(example.get('bboxes', []))\n", " print(f\"\\n--- Displaying File: {filename} | Total Boxes: {total_boxes} ---\")\n", "\n", " except (ValueError, IndexError, TypeError) as e:\n", " print(f\"Error selecting example: {e}\")\n", " return\n", "\n", " # --- IMAGE LOADING (FAST, LOCAL) ---\n", " try:\n", " # Load the image using its filename from the local folder\n", " img = load_local_image(filename)\n", "\n", " except Exception as e:\n", " print(f\"\\nšŸ›‘ Fatal Error: Image for file {filename} could not be loaded from local path. Error: {e}\")\n", " return\n", "\n", " # --- Visualization Setup (Plotting - Unchanged) ---\n", " fig, (ax_img, ax_meta) = plt.subplots(1, 2, figsize=(14, 7), gridspec_kw={'width_ratios': [1, 1]})\n", "\n", " # Left Subplot: Image and Bounding Boxes\n", " ax_img.imshow(img)\n", " bboxes = example.get(\"bboxes\", [])\n", "\n", " img_title = f\"Image: {filename} ({total_boxes} Boxes)\"\n", " ax_img.set_title(img_title, fontsize=TITLE_FONT_SIZE, fontweight='bold')\n", "\n", " # Draw Bounding Boxes\n", " for i, b in enumerate(bboxes[:MAX_BOXES_TO_DRAW]):\n", " x, y, w, h = b[\"x\"], b[\"y\"], b[\"width\"], b[\"height\"]\n", " source = b.get(\"source\", \"N/A\")\n", "\n", " rect = patches.Rectangle((x, y), w, h,\n", " linewidth=2,\n", " edgecolor=BOX_COLOR,\n", " linestyle='-',\n", " fill=False)\n", " ax_img.add_patch(rect)\n", "\n", " ax_img.text(x, y - 5, f\"Source: {source}\",\n", " fontsize=8,\n", " color=TEXT_COLOR,\n", " bbox=dict(facecolor=BOX_LABEL_BG_COLOR, alpha=0.8, edgecolor='none', pad=2))\n", "\n", " ax_img.axis(\"off\")\n", "\n", " # Right Subplot: Structured Metadata\n", " ax_meta.set_axis_off()\n", " ax_meta.set_xlim(0, 1)\n", " ax_meta.set_ylim(0, 1)\n", "\n", " y_pos = 0.95\n", " x_pos = 0.05\n", " line_height = 0.08\n", "\n", " ax_meta.text(x_pos, y_pos, \"Metadata Details\", fontsize=METADATA_SECTION_TITLE_SIZE + 2,\n", " fontweight='bold', transform=ax_meta.transAxes)\n", " y_pos -= line_height * 1.5\n", "\n", " # Caption\n", " caption = example.get(\"caption_text\", \"N/A (No caption available)\")\n", " ax_meta.text(x_pos, y_pos, \"Caption:\", fontsize=METADATA_SECTION_TITLE_SIZE,\n", " fontweight='bold', transform=ax_meta.transAxes)\n", " y_pos -= line_height\n", "\n", " wrapped_caption = \"\\n\".join([caption[i:i+60] for i in range(0, len(caption), 60)])\n", " ax_meta.text(x_pos, y_pos, wrapped_caption, fontsize=METADATA_TEXT_SIZE, transform=ax_meta.transAxes,\n", " verticalalignment='top')\n", " y_pos -= line_height * (wrapped_caption.count('\\n') + 2)\n", "\n", " # Clinical and Publication Details\n", " meta = example.get(\"meta\", {})\n", " ax_meta.text(x_pos, y_pos, \"Clinical & Publication:\", fontsize=METADATA_SECTION_TITLE_SIZE,\n", " fontweight='bold', transform=ax_meta.transAxes)\n", " y_pos -= line_height\n", "\n", " details = [\n", " (\"Final Diagnosis:\", meta.get('final_diagnosis', 'N/A')),\n", " (\"Title:\", meta.get('title', 'N/A')),\n", " (\"Publication Date:\", meta.get('publication_date', 'N/A')),\n", " (\"Link:\", meta.get('link', 'N/A'))\n", " ]\n", "\n", " for label, value in details:\n", " ax_meta.text(x_pos, y_pos, f\"{label:<18} {value}\", fontsize=METADATA_TEXT_SIZE, transform=ax_meta.transAxes)\n", " y_pos -= line_height\n", "\n", " plt.tight_layout()\n", " plt.show()" ], "metadata": { "id": "-QzYNu9dFtG9" }, "execution_count": null, "outputs": [] }, { "cell_type": "code", "source": [ "# --- Demo Examples (Instant after initial caching) ---\n", "\n", "# 1. Default Example: Find the first entry that has bounding boxes\n", "print(\"--- DEMO 1: Displaying first example with boxes ---\")\n", "display_nova_example(ds)\n", "\n", "# 2. Select by Integer Index\n", "print(\"\\n--- DEMO 2: Select by Index (95) ---\")\n", "display_nova_example(ds, selector=543)\n", "\n", "# 3. Select by String Filename\n", "print(\"\\n--- DEMO 3: Select by Filename (Logical ID) ---\")\n", "display_nova_example(ds, selector=ds[0]['filename'])" ], "metadata": { "id": "6O1rffbKwOxV" }, "execution_count": null, "outputs": [] }, { "cell_type": "code", "source": [ "import os\n" ], "metadata": { "id": "zZ8r9w7YFa6Y" }, "execution_count": null, "outputs": [] }, { "cell_type": "code", "source": [], "metadata": { "id": "mPsz_1vZLIFM" }, "execution_count": null, "outputs": [] } ] }