Replace Colab notebooks with clean set aligned to tutorials

- Delete notebooks/241106_colab_JB/ (5 notebooks with branch-specific installs)
- Delete old notebooks/colab/ (5 notebooks with deprecated API calls)
- Add 4 clean notebooks mirroring Sphinx tutorials 01-04:
  01_getting_started: end-to-end tiling, training, prediction
  02_data_preparation: RGB/MS tiling, multi-class, advanced options
  03_training_and_evaluation: registration, config, training, loss plotting
  04_prediction: full postprocessing pipeline including post_clean
- All use `pip install detectree2` (PyPI) and current API
- Source cells only, no outputs (run on Colab to populate)

Co-authored-by: Cursor <cursoragent@cursor.com>

Author: PatBall1

notebooks/241106_colab_JB/tile_multiclass_JB.ipynb

@@ -0,0 +1,265 @@
+{
+ "nbformat": 4,
+ "nbformat_minor": 0,
+ "metadata": {
+  "colab": {
+   "provenance": []
+  },
+  "kernelspec": {
+   "name": "python3",
+   "display_name": "Python 3"
+  },
+  "language_info": {
+   "name": "python"
+  },
+  "accelerator": "GPU"
+ },
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# Getting Started: Your First Prediction\n",
+    "\n",
+    "This notebook provides a concise, end-to-end walkthrough to get you from an orthomosaic to a final crown prediction map using **detectree2**.\n",
+    "\n",
+    "The key steps are:\n",
+    "1. Preparing data (tiling)\n",
+    "2. Training a model\n",
+    "3. Making landscape-level predictions\n",
+    "\n",
+    "For the full tutorial, see the [documentation](https://patball1.github.io/detectree2/tutorials/01_getting_started.html).\n",
+    "\n",
+    "Example data is available on [Zenodo](https://zenodo.org/records/8136161)."
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Setup"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "!pip install torch torchvision torchaudio\n",
+    "!pip install 'git+https://github.com/facebookresearch/detectron2.git'\n",
+    "!pip install detectree2"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from google.colab import drive\n",
+    "drive.mount('/content/drive')"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## 1. Preparing Data\n",
+    "\n",
+    "First, we tile our large orthomosaic and crown data into smaller images suitable for training.\n",
+    "\n",
+    "You will need:\n",
+    "- An orthomosaic (`.tif`)\n",
+    "- Corresponding tree crown polygons (`.gpkg` or `.shp`)\n",
+    "\n",
+    "For best results, manual crowns should be supplied as dense clusters rather than sparsely scattered across the landscape."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from detectree2.preprocessing.tiling import tile_data, to_traintest_folders\n",
+    "import geopandas as gpd\n",
+    "import rasterio"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Set up input paths\n",
+    "site_path = \"./Paracou\"  # Example path\n",
+    "img_path = site_path + \"/rgb/Paracou_RGB_2016_10cm.tif\"\n",
+    "crown_path = site_path + \"/crowns/UpdatedCrowns8.gpkg\"\n",
+    "\n",
+    "# Read in crowns and match CRS to the image\n",
+    "data = rasterio.open(img_path)\n",
+    "crowns = gpd.read_file(crown_path)\n",
+    "crowns = crowns.to_crs(data.crs.data)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Set tiling parameters\n",
+    "buffer = 30\n",
+    "tile_width = 40\n",
+    "tile_height = 40\n",
+    "threshold = 0.6\n",
+    "out_dir = site_path + \"/tiles/\"\n",
+    "\n",
+    "# Tile the data for training\n",
+    "tile_data(img_path, out_dir, buffer, tile_width, tile_height, crowns, threshold, mode=\"rgb\")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Create train/test folders\n",
+    "to_traintest_folders(out_dir, out_dir, test_frac=0.15)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## 2. Training a Model\n",
+    "\n",
+    "Register the training data, configure the model, and train."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from detectree2.models.train import register_train_data, MyTrainer, setup_cfg\n",
+    "\n",
+    "train_location = out_dir + \"/train/\"\n",
+    "register_train_data(train_location, 'Paracou', val_fold=5)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Set the base (pre-trained) model from the detectron2 model_zoo\n",
+    "base_model = \"COCO-InstanceSegmentation/mask_rcnn_R_101_FPN_3x.yaml\"\n",
+    "\n",
+    "trains = (\"Paracou_train\",)  # Registered train data\n",
+    "tests = (\"Paracou_val\",)    # Registered validation data\n",
+    "\n",
+    "model_output_dir = \"./train_outputs\"\n",
+    "\n",
+    "cfg = setup_cfg(base_model, trains, tests, workers=4, eval_period=100, max_iter=3000, out_dir=model_output_dir)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "trainer = MyTrainer(cfg, patience=5)\n",
+    "trainer.resume_or_load(resume=False)\n",
+    "trainer.train()"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## 3. Making Landscape-Level Predictions\n",
+    "\n",
+    "Tile the full orthomosaic, run predictions, then project back to geographic coordinates."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from detectree2.models.predict import predict_on_data\n",
+    "from detectree2.models.outputs import project_to_geojson, stitch_crowns, clean_crowns\n",
+    "from detectron2.engine import DefaultPredictor\n",
+    "\n",
+    "# Path to the full orthomosaic\n",
+    "img_path = site_path + \"/rgb/Paracou_RGB_2016_10cm.tif\"\n",
+    "pred_tiles_path = site_path + \"/tiles_pred/\"\n",
+    "\n",
+    "# Specify tiling parameters (should be similar to training)\n",
+    "buffer = 30\n",
+    "tile_width = 40\n",
+    "tile_height = 40\n",
+    "tile_data(img_path, pred_tiles_path, buffer, tile_width, tile_height)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# You can use your own trained model or download a pre-trained one\n",
+    "# !wget https://zenodo.org/records/15863800/files/250312_flexi.pth\n",
+    "\n",
+    "trained_model = \"./230103_randresize_full.pth\"\n",
+    "cfg = setup_cfg(update_model=trained_model)\n",
+    "predictor = DefaultPredictor(cfg)\n",
+    "predict_on_data(pred_tiles_path, predictor)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Project tile predictions to geo-referenced crowns\n",
+    "project_to_geojson(pred_tiles_path, pred_tiles_path + \"predictions/\", pred_tiles_path + \"predictions_geo/\")\n",
+    "\n",
+    "# Stitch and clean crowns\n",
+    "crowns = stitch_crowns(pred_tiles_path + \"predictions_geo/\")\n",
+    "clean = clean_crowns(crowns, 0.6, confidence=0.5)  # Filter low-confidence and overlapping crowns"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## 4. Saving and Visualizing\n",
+    "\n",
+    "Save the cleaned crown map. You can view the output in QGIS or ArcGIS."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Simplify geometries for easier editing in GIS software\n",
+    "clean = clean.set_geometry(clean.simplify(0.3))\n",
+    "\n",
+    "# Save to file\n",
+    "clean.to_file(site_path + \"/crowns_out.gpkg\", driver=\"GPKG\")"
+   ]
+  }
+ ]
+}