Fast (vectorized) implementation of Octree encoding

encode_octree.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "import math\n",
    "import numpy as np \n",
    "import pandas as pd"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [],
   "source": [
    "nodes = np.zeros(3000000 * 8, np.uint32)\n",
    "bit_values = 2 ** np.arange(8)\n",
    "\n",
    "def encode_octree(points, precision, min_range):\n",
    "    max_level = math.ceil(math.log2(min_range / precision))\n",
    "    num_points = points.shape[0]\n",
    "    \n",
    "    grid_size_0 = precision * 2 ** max_level\n",
    "    grid_size = grid_size_0\n",
    "    \n",
    "    paths = np.zeros((num_points, max_level + 1), np.uint32)\n",
    "    \n",
    "    x = points['x'] + grid_size_0 - precision / 2.0\n",
    "    y = points['y'] + grid_size_0 - precision / 2.0\n",
    "    z = points['z'] + grid_size_0 - precision / 2.0\n",
    "    for j in range(max_level):\n",
    "        x_plus = x >= grid_size\n",
    "        y_plus = y >= grid_size\n",
    "        z_plus = z >= grid_size\n",
    "        paths[:, j] = x_plus * 1 + y_plus * 2 + z_plus * 4\n",
    "        x -= x_plus * grid_size\n",
    "        y -= y_plus * grid_size\n",
    "        z -= z_plus * grid_size\n",
    "        grid_size /= 2.0\n",
    "    x_plus = x >= grid_size\n",
    "    y_plus = y >= grid_size\n",
    "    z_plus = z >= grid_size\n",
    "    paths[:, max_level] = x_plus * 1 + y_plus * 2 + z_plus * 4\n",
    "        \n",
    "    nodes[:8] = 0   \n",
    "    node_counts = [1]\n",
    "    last_count = 1\n",
    "    \n",
    "    current = np.zeros(num_points, np.uint32)\n",
    "    for j in range(max_level):\n",
    "        next_index = current * 8 + paths[:, j]\n",
    "        next_index_set = np.unique(next_index)\n",
    "        node_count = next_index_set.size\n",
    "        node_counts.append(node_count)\n",
    "        nodes[next_index_set] = np.arange(last_count, last_count + node_count)\n",
    "        nodes[(last_count * 8):((last_count + node_count) * 8)] = 0\n",
    "        last_count += node_count\n",
    "        current = nodes[next_index]\n",
    "    next_index = current * 8 + paths[:, max_level]\n",
    "    next_index_set = np.unique(next_index)\n",
    "    leaf_count = next_index_set.size\n",
    "    nodes[next_index_set] = last_count\n",
    "    df['leaf_index'] = next_index\n",
    "    \n",
    "    reshaped_nodes = nodes[:last_count * 8].reshape((-1, 8))\n",
    "    encoded = ((reshaped_nodes != 0) * bit_values).sum(axis = 1)\n",
    "    \n",
    "    # leaf values\n",
    "    grouped = df.groupby('leaf_index')\n",
    "    intensities = grouped['intensity'].max().to_numpy()\n",
    "    \n",
    "    return { 'precision': precision, 'node_counts': node_counts, 'leaf_count': leaf_count }, encoded, intensities"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.9.13"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}