anchor-generator

anchor-generator.ipynb

{
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/gist/ariG23498/777ea321f4c294842f3f7de45dde8258/scratchpad.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "!pip install --upgrade -qq tensorflow\n",
        "!pip install --upgrade -qq keras"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "kXSYq9j_Oq76",
        "outputId": "af0ee3cc-7dca-487a-848f-868479121d89"
      },
      "execution_count": 1,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m475.2/475.2 MB\u001b[0m \u001b[31m2.8 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
            "\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m5.5/5.5 MB\u001b[0m \u001b[31m58.5 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
            "\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m442.0/442.0 kB\u001b[0m \u001b[31m31.1 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
            "\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m1.7/1.7 MB\u001b[0m \u001b[31m58.8 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
            "\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m997.1/997.1 kB\u001b[0m \u001b[31m6.4 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
            "\u001b[?25h\u001b[31mERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.\n",
            "tensorflow 2.15.0 requires keras<2.16,>=2.15.0, but you have keras 3.0.0 which is incompatible.\u001b[0m\u001b[31m\n",
            "\u001b[0m"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "import os\n",
        "os.environ[\"KERAS_BACKEND\"] = \"numpy\"\n",
        "from keras import ops\n",
        "import math"
      ],
      "metadata": {
        "id": "VJJSDbHOWvFM"
      },
      "execution_count": 2,
      "outputs": []
    },
    {
      "cell_type": "code",
      "execution_count": 3,
      "metadata": {
        "id": "lIYdn1woOS1n"
      },
      "outputs": [],
      "source": [
        "sizes = [2**x for x in [5, 6, 7]]\n",
        "scales = [2**x for x in [0, 1/3, 2/3]]\n",
        "aspect_ratios = [1/2, 1/1, 2/1]\n",
        "strides = [2**x for x in [3, 4, 5]]"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "def param_to_level_dict(param):\n",
        "    result = dict()\n",
        "    for i in range(len(param)):\n",
        "        result[i] = param[i]\n",
        "    return result"
      ],
      "metadata": {
        "id": "P11eSroq_CMz"
      },
      "execution_count": 4,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "sizes = param_to_level_dict(sizes)\n",
        "strides = param_to_level_dict(strides)"
      ],
      "metadata": {
        "id": "HOsU26mdBHN9"
      },
      "execution_count": 5,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "sizes"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "oWM0jrSKBPMJ",
        "outputId": "cc106551-9802-47c3-f76e-f77c0029bf0e"
      },
      "execution_count": 6,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "{0: 32, 1: 64, 2: 128}"
            ]
          },
          "metadata": {},
          "execution_count": 6
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "strides"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "7IRvRRRWBPvq",
        "outputId": "6b162d57-c109-48dd-eae6-5bf3f69d1938"
      },
      "execution_count": 7,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "{0: 8, 1: 16, 2: 32}"
            ]
          },
          "metadata": {},
          "execution_count": 7
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "def match_param_structure_to_sizes(params, sizes):\n",
        "    return {key: params for key in sizes.keys()}"
      ],
      "metadata": {
        "id": "5-LdF9wcBQ1i"
      },
      "execution_count": 8,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "aspect_ratios = match_param_structure_to_sizes(\n",
        "    aspect_ratios, sizes\n",
        ")\n",
        "scales = match_param_structure_to_sizes(\n",
        "    scales, sizes\n",
        ")"
      ],
      "metadata": {
        "id": "pgY6sI1SBZ_t"
      },
      "execution_count": 9,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "aspect_ratios"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "QUZ1rKnZBhbE",
        "outputId": "03bda8e6-783e-4c82-d373-4b455988922f"
      },
      "execution_count": 10,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "{0: [0.5, 1.0, 2.0], 1: [0.5, 1.0, 2.0], 2: [0.5, 1.0, 2.0]}"
            ]
          },
          "metadata": {},
          "execution_count": 10
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "scales"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "hJfIfUcdBii5",
        "outputId": "2af1338c-865a-4c3a-a0e2-4474c4458a5c"
      },
      "execution_count": 11,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "{0: [1, 1.2599210498948732, 1.5874010519681994],\n",
              " 1: [1, 1.2599210498948732, 1.5874010519681994],\n",
              " 2: [1, 1.2599210498948732, 1.5874010519681994]}"
            ]
          },
          "metadata": {},
          "execution_count": 11
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "anchor_generators = {}\n",
        "for k in sizes.keys():\n",
        "    anchor_generators[k] = {\n",
        "        \"sizes\": sizes[k],\n",
        "        \"scales\": scales[k],\n",
        "        \"aspect_ratios\": aspect_ratios[k],\n",
        "        \"strides\": strides[k],\n",
        "    }"
      ],
      "metadata": {
        "id": "KK48MXtCBz0z"
      },
      "execution_count": 12,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "anchor_generators"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "yHVQt-98CMZ5",
        "outputId": "103353cf-9429-46f6-9f67-62885be09688"
      },
      "execution_count": 13,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "{0: {'sizes': 32,\n",
              "  'scales': [1, 1.2599210498948732, 1.5874010519681994],\n",
              "  'aspect_ratios': [0.5, 1.0, 2.0],\n",
              "  'strides': 8},\n",
              " 1: {'sizes': 64,\n",
              "  'scales': [1, 1.2599210498948732, 1.5874010519681994],\n",
              "  'aspect_ratios': [0.5, 1.0, 2.0],\n",
              "  'strides': 16},\n",
              " 2: {'sizes': 128,\n",
              "  'scales': [1, 1.2599210498948732, 1.5874010519681994],\n",
              "  'aspect_ratios': [0.5, 1.0, 2.0],\n",
              "  'strides': 32}}"
            ]
          },
          "metadata": {},
          "execution_count": 13
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "```\n",
        "SingleAnchorGenerator(\n",
        "    sizes=32,\n",
        "    scales=[1, 1.25, 1.58],\n",
        "    aspect_ratios=[0.5, 1.0, 2.0],\n",
        "    stride=8,\n",
        ")\n",
        "```"
      ],
      "metadata": {
        "id": "8t9o93jMNpq7"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "def single_anchor_gen(config):\n",
        "    sizes = config['sizes'] # 32\n",
        "    scales = config['scales'] # [1, 1.25, 1.33]\n",
        "    aspect_ratios = config['aspect_ratios'] # [0.5, 1.0, 2.0]\n",
        "    stride = config['strides'] # 8\n",
        "\n",
        "    image_height = 512\n",
        "    image_width = 512\n",
        "\n",
        "    aspect_ratios = ops.cast(aspect_ratios, \"float32\")\n",
        "    aspect_ratios_sqrt = ops.cast(ops.sqrt(aspect_ratios), dtype=\"float32\")\n",
        "    anchor_size = ops.cast(sizes, \"float32\")\n",
        "\n",
        "    # [K]\n",
        "    anchor_heights = []\n",
        "    anchor_widths = []\n",
        "    for scale in scales:\n",
        "        anchor_size_t = anchor_size * scale\n",
        "        anchor_height = anchor_size_t / aspect_ratios_sqrt\n",
        "        anchor_width = anchor_size_t * aspect_ratios_sqrt\n",
        "        anchor_heights.append(anchor_height)\n",
        "        anchor_widths.append(anchor_width)\n",
        "    anchor_heights = ops.concatenate(anchor_heights, axis=0)\n",
        "    anchor_widths = ops.concatenate(anchor_widths, axis=0)\n",
        "    half_anchor_heights = ops.reshape(0.5 * anchor_heights, [1, 1, -1])\n",
        "    half_anchor_widths = ops.reshape(0.5 * anchor_widths, [1, 1, -1])\n",
        "\n",
        "    stride = stride\n",
        "    # make sure range of `cx` is within limit of `image_width` with\n",
        "    # `stride`, also for sizes where `image_width % stride != 0`.\n",
        "    # [W]\n",
        "    cx = ops.cast(\n",
        "        ops.arange(\n",
        "            0.5 * stride, math.ceil(image_width / stride) * stride, stride\n",
        "        ),\n",
        "        \"float32\",\n",
        "    )\n",
        "    # make sure range of `cy` is within limit of `image_height` with\n",
        "    # `stride`, also for sizes where `image_height % stride != 0`.\n",
        "    # [H]\n",
        "    cy = ops.cast(\n",
        "        ops.arange(\n",
        "            0.5 * stride, math.ceil(image_height / stride) * stride, stride\n",
        "        ),\n",
        "        \"float32\",\n",
        "    )\n",
        "    # [H, W]\n",
        "    cx_grid, cy_grid = ops.meshgrid(cx, cy)\n",
        "    # [H, W, 1]\n",
        "    cx_grid = ops.expand_dims(cx_grid, axis=-1)\n",
        "    cy_grid = ops.expand_dims(cy_grid, axis=-1)\n",
        "\n",
        "    y_min = ops.reshape(cy_grid - half_anchor_heights, (-1,))\n",
        "    y_max = ops.reshape(cy_grid + half_anchor_heights, (-1,))\n",
        "    x_min = ops.reshape(cx_grid - half_anchor_widths, (-1,))\n",
        "    x_max = ops.reshape(cx_grid + half_anchor_widths, (-1,))\n",
        "\n",
        "    # [H * W * K, 1]\n",
        "    y_min = ops.expand_dims(y_min, axis=-1)\n",
        "    y_max = ops.expand_dims(y_max, axis=-1)\n",
        "    x_min = ops.expand_dims(x_min, axis=-1)\n",
        "    x_max = ops.expand_dims(x_max, axis=-1)\n",
        "\n",
        "    # [H * W * K, 4]\n",
        "    return ops.cast(\n",
        "        ops.concatenate([y_min, x_min, y_max, x_max], axis=-1), \"float32\"\n",
        "    )"
      ],
      "metadata": {
        "id": "MvGGb4jQCN-3"
      },
      "execution_count": 23,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "single_anchor_gen(anchor_generators[0])"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "4FYWI2dnWdma",
        "outputId": "089c223f-8b91-409d-a47a-8e779bac48a3"
      },
      "execution_count": 24,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "array([[-18.627417 ,  -7.3137083,  26.627417 ,  15.313708 ],\n",
              "       [-12.       , -12.       ,  20.       ,  20.       ],\n",
              "       [ -7.3137083, -18.627417 ,  15.313708 ,  26.627417 ],\n",
              "       ...,\n",
              "       [472.0812   , 490.04062  , 543.91876  , 525.9594   ],\n",
              "       [482.6016   , 482.6016   , 533.39844  , 533.39844  ],\n",
              "       [490.04062  , 472.0812   , 525.9594   , 543.91876  ]],\n",
              "      dtype=float32)"
            ]
          },
          "metadata": {},
          "execution_count": 24
        }
      ]
    }
  ],
  "metadata": {
    "colab": {
      "name": "scratchpad",
      "provenance": [],
      "include_colab_link": true
    },
    "kernelspec": {
      "display_name": "Python 3",
      "name": "python3"
    }
  },
  "nbformat": 4,
  "nbformat_minor": 0
}