notebooks/GradientVisual.ipynb

{
  "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": "99NhDXLp8paJ"
      },
      "outputs": [],
      "source": [
        "import collections\n",
        "tf_counter = collections.Counter({(0, 0, 0): 1822737431,\n",
        "         (255, 255, 255): 1842426210,\n",
        "         (128, 128, 128): 714368785,\n",
        "         (220, 220, 220): 10689714,\n",
        "         (127, 127, 127): 56486751,\n",
        "         (110, 110, 110): 197108571,\n",
        "         (145, 145, 145): 19794432,\n",
        "         (164, 164, 164): 49428,\n",
        "         (155, 155, 155): 167247,\n",
        "         (146, 146, 146): 50574,\n",
        "         (137, 137, 137): 64224,\n",
        "         (92, 92, 92): 49388,\n",
        "         (101, 101, 101): 169829,\n",
        "         (119, 119, 119): 64157,\n",
        "         (240, 240, 240): 218237258,\n",
        "         (173, 173, 173): 473,\n",
        "         (84, 84, 84): 478,\n",
        "         (66, 66, 66): 27916988,\n",
        "         (32, 32, 32): 9959936,\n",
        "         (192, 192, 192): 2956576,\n",
        "         (96, 96, 96): 10010112,\n",
        "         (243, 243, 243): 28654651,\n",
        "         (188, 188, 188): 3111760,\n",
        "         (80, 80, 80): 492480,\n",
        "         (70, 70, 70): 6840,\n",
        "         (15, 15, 15): 19869696,\n",
        "         (209, 209, 209): 10171364,\n",
        "         (98, 98, 98): 17449600,\n",
        "         (187, 187, 187): 2626400,\n",
        "         (213, 213, 213): 54800,\n",
        "         (201, 201, 201): 48400,\n",
        "         (215, 215, 215): 57587,\n",
        "         (198, 198, 198): 447600,\n",
        "         (199, 199, 199): 306080,\n",
        "         (197, 197, 197): 147200,\n",
        "         (218, 218, 218): 36105,\n",
        "         (225, 225, 225): 26197,\n",
        "         (204, 204, 204): 45520,\n",
        "         (222, 222, 222): 35710,\n",
        "         (196, 196, 196): 19200,\n",
        "         (207, 207, 207): 72520,\n",
        "         (219, 219, 219): 47589,\n",
        "         (216, 216, 216): 17484,\n",
        "         (202, 202, 202): 47840,\n",
        "         (224, 224, 224): 26880,\n",
        "         (227, 227, 227): 9220,\n",
        "         (208, 208, 208): 37280,\n",
        "         (214, 214, 214): 40204,\n",
        "         (232, 232, 232): 2583,\n",
        "         (200, 200, 200): 44640,\n",
        "         (212, 212, 212): 19200,\n",
        "         (221, 221, 221): 26341,\n",
        "         (229, 229, 229): 2695,\n",
        "         (211, 211, 211): 58227,\n",
        "         (205, 205, 205): 27600,\n",
        "         (203, 203, 203): 35120,\n",
        "         (230, 230, 230): 5547,\n",
        "         (210, 210, 210): 39620,\n",
        "         (233, 233, 233): 9389,\n",
        "         (226, 226, 226): 12044,\n",
        "         (234, 234, 234): 4460,\n",
        "         (217, 217, 217): 24355,\n",
        "         (223, 223, 223): 30243,\n",
        "         (228, 228, 228): 9355,\n",
        "         (206, 206, 206): 27280,\n",
        "         (231, 231, 231): 2668,\n",
        "         (239, 239, 239): 1806,\n",
        "         (237, 237, 237): 2058})"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "def gradient_value(c, granularity = 224 ** 2):\n",
        "  out = []\n",
        "  n = c.total() // granularity\n",
        "  i = 0\n",
        "  while len(out) < n:\n",
        "    count = 0\n",
        "    total = 0\n",
        "    while count < granularity:\n",
        "      diff = min(c[i, i, i], granularity - count)\n",
        "      count += diff\n",
        "      total += i * diff\n",
        "      c[i, i, i] -= diff\n",
        "      if not c[i, i, i]:\n",
        "        i += 1\n",
        "    out.append(total / count)\n",
        "  return out\n",
        "\n",
        "gradient = gradient_value(collections.Counter(tf_counter), 112 ** 2)"
      ],
      "metadata": {
        "id": "hzdYzrVZ9bFb"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "len(gradient)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "Hn_7EIk59dvU",
        "outputId": "9d8b189a-7b10-4f84-d561-5ab2b3c68319"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "400000"
            ]
          },
          "metadata": {},
          "execution_count": 3
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "col = 200\n",
        "b = [gradient[i:i + col] for i in range(0, len(gradient), col)]\n",
        "b = list(zip(*b))"
      ],
      "metadata": {
        "id": "3vQSEgck__PC"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "len(b)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "b9qqUm58ApHu",
        "outputId": "4241e0ff-d8b3-43c5-f9c4-10ac9eb649e5"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "200"
            ]
          },
          "metadata": {},
          "execution_count": 5
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "len(b[0])"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "gUztNFZ-BHvQ",
        "outputId": "19f2e77c-71a7-44e5-b933-bf73ff264927"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "2000"
            ]
          },
          "metadata": {},
          "execution_count": 6
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "from PIL import Image\n",
        "import numpy as np\n",
        "Image.fromarray(np.array(b)).convert('L')"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 177
        },
        "id": "okC3TU2OBRhG",
        "outputId": "1e8f6b20-4893-41b1-b886-a9a634cc1456"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<PIL.Image.Image image mode=L size=2000x200>"
            ],
            "image/png": "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\n"
          },
          "metadata": {},
          "execution_count": 7
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "torch_mid_fill_counter = collections.Counter({(128, 128, 128): 283712442,\n",
        "         (0, 0, 0): 2490944591,\n",
        "         (255, 255, 255): 1358923850,\n",
        "         (127, 127, 127): 11714748,\n",
        "         (38, 38, 38): 114028300,\n",
        "         (216, 216, 216): 107856960,\n",
        "         (17, 17, 17): 6362113,\n",
        "         (237, 237, 237): 5572814,\n",
        "         (23, 23, 23): 183391,\n",
        "         (231, 231, 231): 4664861,\n",
        "         (254, 254, 254): 364276458,\n",
        "         (236, 236, 236): 1017360,\n",
        "         (230, 230, 230): 14130,\n",
        "         (178, 178, 178): 112517216,\n",
        "         (89, 89, 89): 5880312,\n",
        "         (1, 1, 1): 19587968,\n",
        "         (246, 246, 246): 133196,\n",
        "         (136, 136, 136): 129836,\n",
        "         (234, 234, 234): 3406,\n",
        "         (131, 131, 131): 33660,\n",
        "         (20, 20, 20): 3441,\n",
        "         (125, 125, 125): 33707,\n",
        "         (8, 8, 8): 133406,\n",
        "         (119, 119, 119): 261794,\n",
        "         (37, 37, 37): 19169012,\n",
        "         (5, 5, 5): 105687,\n",
        "         (243, 243, 243): 110932,\n",
        "         (252, 252, 252): 16866,\n",
        "         (249, 249, 249): 108333,\n",
        "         (133, 133, 133): 1316,\n",
        "         (122, 122, 122): 29351,\n",
        "         (116, 116, 116): 25790,\n",
        "         (11, 11, 11): 4368340,\n",
        "         (2, 2, 2): 207308,\n",
        "         (14, 14, 14): 512,\n",
        "         (134, 134, 134): 28082,\n",
        "         (228, 228, 228): 170288,\n",
        "         (139, 139, 139): 25790,\n",
        "         (26, 26, 26): 5871104,\n",
        "         (240, 240, 240): 512,\n",
        "         (225, 225, 225): 1865,\n",
        "         (29, 29, 29): 4098,\n",
        "         (39, 39, 39): 6096384,\n",
        "         (191, 191, 191): 7406784,\n",
        "         (166, 166, 166): 5779704,\n",
        "         (16, 16, 16): 530460,\n",
        "         (22, 22, 22): 170604,\n",
        "         (137, 137, 137): 131300,\n",
        "         (221, 221, 221): 2007780,\n",
        "         (177, 177, 177): 11716096,\n",
        "         (232, 232, 232): 1258866,\n",
        "         (12, 12, 12): 160704,\n",
        "         (165, 165, 165): 308448,\n",
        "         (161, 161, 161): 4284,\n",
        "         (18, 18, 18): 298080,\n",
        "         (24, 24, 24): 4140,\n",
        "         (190, 190, 190): 164592,\n",
        "         (194, 194, 194): 2286,\n",
        "         (77, 77, 77): 6573056,\n",
        "         (215, 215, 215): 12769792,\n",
        "         (242, 242, 242): 4114432,\n",
        "         (25, 25, 25): 169776,\n",
        "         (21, 21, 21): 2358,\n",
        "         (204, 204, 204): 3537408,\n",
        "         (151, 151, 151): 5870592,\n",
        "         (244, 244, 244): 342954,\n",
        "         (31, 31, 31): 150336,\n",
        "         (222, 222, 222): 150336,\n",
        "         (253, 253, 253): 10104704,\n",
        "         (41, 41, 41): 2088,\n",
        "         (212, 212, 212): 2088,\n",
        "         (217, 217, 217): 6471664,\n",
        "         (50, 50, 50): 154224,\n",
        "         (203, 203, 203): 321408,\n",
        "         (54, 54, 54): 4464,\n",
        "         (199, 199, 199): 4464,\n",
        "         (64, 64, 64): 2935296,\n",
        "         (189, 189, 189): 2935296,\n",
        "         (251, 251, 251): 190512,\n",
        "         (4, 4, 4): 2646,\n",
        "         (49, 49, 49): 3227920,\n",
        "         (124, 124, 124): 3286528})"
      ],
      "metadata": {
        "id": "78MLfba2Bblm"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "gradient = gradient_value(collections.Counter(torch_mid_fill_counter), 112 ** 2)\n",
        "b = [gradient[i:i + col] for i in range(0, len(gradient), col)]\n",
        "b = list(zip(*b))\n",
        "Image.fromarray(np.array(b)).convert('L')"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 177
        },
        "id": "RZYlrYB_B-KY",
        "outputId": "f6a2b4bc-a4ad-4137-e080-70460b0a9494"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<PIL.Image.Image image mode=L size=2000x200>"
            ],
            "image/png": "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\n"
          },
          "metadata": {},
          "execution_count": 9
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "torch_counter = collections.Counter({(0, 0, 0): 2805231902,\n",
        "         (255, 255, 255): 1358486472,\n",
        "         (254, 254, 254): 365481466,\n",
        "         (177, 177, 177): 12819968,\n",
        "         (178, 178, 178): 114701526,\n",
        "         (38, 38, 38): 88778956,\n",
        "         (216, 216, 216): 83030036,\n",
        "         (17, 17, 17): 6524954,\n",
        "         (237, 237, 237): 5644633,\n",
        "         (23, 23, 23): 190534,\n",
        "         (231, 231, 231): 3600734,\n",
        "         (35, 35, 35): 14253876,\n",
        "         (214, 214, 214): 6256320,\n",
        "         (25, 25, 25): 202176,\n",
        "         (228, 228, 228): 202176,\n",
        "         (21, 21, 21): 2808,\n",
        "         (232, 232, 232): 2808,\n",
        "         (26, 26, 26): 6447616,\n",
        "         (151, 151, 151): 6447616,\n",
        "         (32, 32, 32): 13442128,\n",
        "         (210, 210, 210): 9990064,\n",
        "         (49, 49, 49): 3310128,\n",
        "         (2, 2, 2): 185022,\n",
        "         (251, 251, 251): 182736,\n",
        "         (4, 4, 4): 2538,\n",
        "         (249, 249, 249): 102429,\n",
        "         (236, 236, 236): 1110672,\n",
        "         (230, 230, 230): 15426,\n",
        "         (1, 1, 1): 19626784,\n",
        "         (124, 124, 124): 2634240,\n",
        "         (5, 5, 5): 104423,\n",
        "         (243, 243, 243): 104138,\n",
        "         (11, 11, 11): 3418332,\n",
        "         (225, 225, 225): 5584,\n",
        "         (29, 29, 29): 4313,\n",
        "         (190, 190, 190): 164592,\n",
        "         (194, 194, 194): 2286,\n",
        "         (31, 31, 31): 164592,\n",
        "         (222, 222, 222): 164592,\n",
        "         (253, 253, 253): 9240352,\n",
        "         (252, 252, 252): 2286,\n",
        "         (41, 41, 41): 2286,\n",
        "         (212, 212, 212): 2286,\n",
        "         (203, 203, 203): 317520,\n",
        "         (54, 54, 54): 4410,\n",
        "         (199, 199, 199): 4410,\n",
        "         (18, 18, 18): 301968,\n",
        "         (24, 24, 24): 4194,\n",
        "         (242, 242, 242): 3161088,\n",
        "         (36, 36, 36): 6923448,\n",
        "         (215, 215, 215): 20497736,\n",
        "         (37, 37, 37): 14074368,\n",
        "         (16, 16, 16): 534384,\n",
        "         (22, 22, 22): 191394,\n",
        "         (165, 165, 165): 311040,\n",
        "         (161, 161, 161): 4320,\n",
        "         (12, 12, 12): 160704,\n",
        "         (219, 219, 219): 474,\n",
        "         (213, 213, 213): 6183436,\n",
        "         (39, 39, 39): 5745152,\n",
        "         (50, 50, 50): 168480,\n",
        "         (64, 64, 64): 3035648,\n",
        "         (189, 189, 189): 3035648,\n",
        "         (77, 77, 77): 6547968,\n",
        "         (244, 244, 244): 365292,\n",
        "         (204, 204, 204): 3738112})"
      ],
      "metadata": {
        "id": "zz2YHVH8C-bM"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "gradient = gradient_value(collections.Counter(torch_counter), 112 ** 2)\n",
        "b = [gradient[i:i + col] for i in range(0, len(gradient), col)]\n",
        "b = list(zip(*b))\n",
        "Image.fromarray(np.array(b)).convert('L')"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 177
        },
        "id": "kMzkpoA6F230",
        "outputId": "c3344053-b580-414b-8f27-d2b6190bd543"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<PIL.Image.Image image mode=L size=2000x200>"
            ],
            "image/png": "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\n"
          },
          "metadata": {},
          "execution_count": 11
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "row = [0] * 56 + [1] * 56\n",
        "row *= 2\n",
        "row = [255 * b for b in row]\n",
        "tile = [row for _ in range(56)]\n",
        "anti_row = [255 - b for b in row]\n",
        "anti_tile = [anti_row for _ in range(56)]\n",
        "tile.extend(anti_tile)\n",
        "tile *= 2\n",
        "\n",
        "Image.fromarray(np.array(tile, dtype=np.uint8))"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 241
        },
        "id": "_Ymsa52aF5bE",
        "outputId": "411d09fd-2f7d-43b9-a0bc-30d1f591416e"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<PIL.Image.Image image mode=L size=224x224>"
            ],
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAOAAAADgCAAAAAA/RjU9AAABIklEQVR4nO3dsQnAMAwAwSj775zsoCL4wl8vxONaeK6lZzk3H++7l3OMAnUF6grUFagrUFegrkBdgboCdQXqCtQVqCtQV6CuQF2BugJ1BeoK1BWoK1BXoK5AXYG6AnUF6grUjXL/sN33+xcsUFegrkBdgboCdQXqCtQVqCtQV6CuQF2BugJ1BeoK1BWoK1BXoK5AXYG6AnUF6grUFagrUFegrkDd9hyh/yZOUaCuQF2BugJ1BeoK1BWoK1BXoK5AXYG6AnUF6grUFagrUFegrkBdgboCdQXqCtQVqCtQV6CuQF2Buv6b0BWoK1BXoK5AXYG6AnUF6grUFagrUFegrkBdgboCdQXqCtQVqCtQV6CuQF2BugJ1BeoK1BWoK1BXoO4FtcIJu/1QzgEAAAAASUVORK5CYII=\n"
          },
          "metadata": {},
          "execution_count": 12
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "import matplotlib.pyplot as plt\n",
        "import torch\n",
        "\n",
        "torch_tile = torch.Tensor([tile] * 3).to(torch.uint8)\n",
        "plt.imshow(torch_tile.permute(1, 2, 0))"
      ],
      "metadata": {
        "id": "J3TqFx4yHivU",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 452
        },
        "outputId": "893498b6-2fe4-4a47-c677-8a1c4e099c5f"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<matplotlib.image.AxesImage at 0x7cdbb890bb80>"
            ]
          },
          "metadata": {},
          "execution_count": 14
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 640x480 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "from torchvision.transforms import v2\n",
        "\n",
        "import math\n",
        "from typing import Any, Callable, cast, Dict, List, Optional, Tuple, Type, Union\n",
        "\n",
        "import PIL.Image\n",
        "import torch\n",
        "\n",
        "from torch.utils._pytree import tree_flatten, tree_unflatten, TreeSpec\n",
        "from torchvision import transforms as _transforms, tv_tensors\n",
        "from torchvision.transforms import _functional_tensor as _FT\n",
        "from torchvision.transforms.v2 import AutoAugmentPolicy, functional as F, InterpolationMode, Transform\n",
        "from torchvision.transforms.v2.functional._geometry import _check_interpolation\n",
        "from torchvision.transforms.v2.functional._meta import get_size\n",
        "from torchvision.transforms.v2.functional._utils import _FillType, _FillTypeJIT\n",
        "\n",
        "from torchvision.transforms.v2._utils import _get_fill, _setup_fill_arg, check_type, is_pure_tensor\n",
        "\n",
        "ImageOrVideo = Union[torch.Tensor, PIL.Image.Image, tv_tensors.Image, tv_tensors.Video]\n",
        "\n",
        "\n",
        "def _solarize_add(\n",
        "    image: ImageOrVideo, addition: int = 0, threshold: int = 128\n",
        ") -> ImageOrVideo:\n",
        "    bound = _FT._max_value(image.dtype) if isinstance(image, torch.Tensor) else 255\n",
        "    added_image = image.to(torch.int64) + addition\n",
        "    added_image = added_image.clip(0, bound).to(torch.uint8)\n",
        "    return torch.where(image < threshold, added_image, image)\n",
        "\n",
        "\n",
        "def _cutout(\n",
        "    image: ImageOrVideo,\n",
        "    pad_size: int,\n",
        "    replace: int = 0,\n",
        ") -> ImageOrVideo:\n",
        "    _, img_h, img_w = F.get_dimensions(image)\n",
        "\n",
        "    # Sample the center location in the image where the zero mask will be applied.\n",
        "    cutout_center_height = int(torch.randint(img_h, ()))\n",
        "    cutout_center_width = int(torch.randint(img_w, ()))\n",
        "\n",
        "    lower_pad = max(0, cutout_center_height - pad_size)\n",
        "    upper_pad = max(0, img_h - cutout_center_height - pad_size)\n",
        "    left_pad = max(0, cutout_center_width - pad_size)\n",
        "    right_pad = max(0, img_w - cutout_center_width - pad_size)\n",
        "\n",
        "    cutout_shape = [img_h - (lower_pad + upper_pad), img_w - (left_pad + right_pad)]\n",
        "    return F.erase(image, lower_pad, left_pad, cutout_shape[0], cutout_shape[1], torch.tensor(replace).unsqueeze(1).unsqueeze(1))\n",
        "\n",
        "\n",
        "class RandAugment17(v2.RandAugment):\n",
        "    def _apply_image_or_video_transform(\n",
        "        self,\n",
        "        image: ImageOrVideo,\n",
        "        transform_id: str,\n",
        "        magnitude: float,\n",
        "        interpolation: Union[InterpolationMode, int],\n",
        "        fill: Dict[Union[Type, str], _FillTypeJIT],\n",
        "    ) -> ImageOrVideo:\n",
        "        # Note: this cast is wrong and is only here to make mypy happy (it disagrees with torchscript)\n",
        "        image = cast(torch.Tensor, image)\n",
        "        fill_ = _get_fill(fill, type(image))\n",
        "\n",
        "        if transform_id == \"Identity\":\n",
        "            return image\n",
        "        elif transform_id == \"ShearX\":\n",
        "            # magnitude should be arctan(magnitude)\n",
        "            # official autoaug: (1, level, 0, 0, 1, 0)\n",
        "            # https://github.com/tensorflow/models/blob/dd02069717128186b88afa8d857ce57d17957f03/research/autoaugment/augmentation_transforms.py#L290\n",
        "            # compared to\n",
        "            # torchvision:      (1, tan(level), 0, 0, 1, 0)\n",
        "            # https://github.com/pytorch/vision/blob/0c2373d0bba3499e95776e7936e207d8a1676e65/torchvision/transforms/functional.py#L976\n",
        "            return F.affine(\n",
        "                image,\n",
        "                angle=0.0,\n",
        "                translate=[0, 0],\n",
        "                scale=1.0,\n",
        "                shear=[math.degrees(math.atan(magnitude)), 0.0],\n",
        "                interpolation=interpolation,\n",
        "                fill=fill_,\n",
        "                center=[0, 0],\n",
        "            )\n",
        "        elif transform_id == \"ShearY\":\n",
        "            # magnitude should be arctan(magnitude)\n",
        "            # See above\n",
        "            return F.affine(\n",
        "                image,\n",
        "                angle=0.0,\n",
        "                translate=[0, 0],\n",
        "                scale=1.0,\n",
        "                shear=[0.0, math.degrees(math.atan(magnitude))],\n",
        "                interpolation=interpolation,\n",
        "                fill=fill_,\n",
        "                center=[0, 0],\n",
        "            )\n",
        "        elif transform_id == \"TranslateX\":\n",
        "            return F.affine(\n",
        "                image,\n",
        "                angle=0.0,\n",
        "                translate=[int(magnitude), 0],\n",
        "                scale=1.0,\n",
        "                interpolation=interpolation,\n",
        "                shear=[0.0, 0.0],\n",
        "                fill=fill_,\n",
        "            )\n",
        "        elif transform_id == \"TranslateY\":\n",
        "            return F.affine(\n",
        "                image,\n",
        "                angle=0.0,\n",
        "                translate=[0, int(magnitude)],\n",
        "                scale=1.0,\n",
        "                interpolation=interpolation,\n",
        "                shear=[0.0, 0.0],\n",
        "                fill=fill_,\n",
        "            )\n",
        "        elif transform_id == \"Rotate\":\n",
        "            return F.rotate(image, angle=magnitude, interpolation=interpolation, fill=fill_)\n",
        "        elif transform_id == \"Brightness\":\n",
        "            return F.adjust_brightness(image, brightness_factor=1.0 + magnitude)\n",
        "        elif transform_id == \"Color\":\n",
        "            return F.adjust_saturation(image, saturation_factor=1.0 + magnitude)\n",
        "        elif transform_id == \"Contrast\":\n",
        "            return F.adjust_contrast(image, contrast_factor=1.0 + magnitude)\n",
        "        elif transform_id == \"Sharpness\":\n",
        "            return F.adjust_sharpness(image, sharpness_factor=1.0 + magnitude)\n",
        "        elif transform_id == \"Posterize\":\n",
        "            return F.posterize(image, bits=int(magnitude))\n",
        "        elif transform_id == \"Solarize\":\n",
        "            bound = _FT._max_value(image.dtype) if isinstance(image, torch.Tensor) else 255.0\n",
        "            return F.solarize(image, threshold=bound * magnitude)\n",
        "        elif transform_id == \"AutoContrast\":\n",
        "            return F.autocontrast(image)\n",
        "        elif transform_id == \"Equalize\":\n",
        "            return F.equalize(image)\n",
        "        elif transform_id == \"Invert\":\n",
        "            return F.invert(image)\n",
        "        elif transform_id == \"SolarizeAdd\":\n",
        "            return _solarize_add(image, addition=int(magnitude))\n",
        "        elif transform_id == \"Cutout\":\n",
        "            return _cutout(image, pad_size=int(magnitude), replace=fill_)\n",
        "        else:\n",
        "            raise ValueError(f\"No transform available for {transform_id}\")\n",
        "\n",
        "\n",
        "cutout_const = 40\n",
        "translate_const = 100\n",
        "MAX_LEVEL = 10\n",
        "\n",
        "translate_magnitude = lambda num_bins, _h, _w: torch.linspace(0.0, translate_const, num_bins)\n",
        "shear_magnitude = lambda num_bins, _h, _w: torch.linspace(0.0, 0.3, num_bins)\n",
        "enhance_magnitude = lambda num_bins, _h, _w: torch.linspace(0, 0.9, num_bins)  # It was -0.9, 0.9 but negative magnitude results in the opposite effect\n",
        "\n",
        "RandAugment17._AUGMENTATION_SPACE = {\n",
        "    \"TranslateX\": (translate_magnitude, True),\n",
        "    \"TranslateY\": (translate_magnitude, True),\n",
        "    \"ShearX\": (shear_magnitude, True),\n",
        "    \"ShearY\": (shear_magnitude, True),\n",
        "    \"Rotate\": (lambda num_bins, height, width: torch.linspace(0.0, 30.0, num_bins), True),\n",
        "    \"Brightness\": (enhance_magnitude, False),\n",
        "    \"Color\": (enhance_magnitude, False),\n",
        "    \"Contrast\": (enhance_magnitude, False),\n",
        "    \"Sharpness\": (enhance_magnitude, False),\n",
        "    \"Posterize\": (  # Unchanged!\n",
        "        lambda num_bins, height, width: (8 - (torch.arange(num_bins) / ((num_bins - 1) / 4))).round().int(),\n",
        "        False,\n",
        "    ),\n",
        "    \"Solarize\": (lambda num_bins, height, width: torch.linspace(1.0, 0.0, num_bins), False),  # Unchanged!\n",
        "    \"AutoContrast\": (lambda num_bins, height, width: None, False),  # Unchanged!\n",
        "    \"Equalize\": (lambda num_bins, height, width: None, False),  # Unchanged!\n",
        "    \"Invert\": (lambda num_bins, height, width: None, False),  # \"New\" (equivalent to MAX_LEVEL Solarize)!\n",
        "    \"SolarizeAdd\": (lambda num_bins, height, width: torch.linspace(0., 110., num_bins), False),  # New!\n",
        "    \"Cutout\": (lambda num_bins, height, width: torch.linspace(0., float(cutout_const), num_bins), False),  # New!\n",
        "    # https://github.com/pytorch/vision/pull/6609\n",
        "}\n",
        "transform = RandAugment17(2, 10, num_magnitude_bins=MAX_LEVEL + 1, fill=[128] * 3)"
      ],
      "metadata": {
        "id": "9c3puogDanEG"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "import tqdm\n",
        "N = 100000\n",
        "torch_counter = collections.Counter()\n",
        "for _ in tqdm.tqdm(range(N)):\n",
        "  out = transform(torch_tile)\n",
        "  torch_counter.update(tuple(pixel) for row in out.permute(1, 2, 0).numpy() for pixel in row)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "EkdqLWq-at4N",
        "outputId": "555eac15-3d65-4d70-e068-bff25f41fa6e"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "100%|██████████| 100000/100000 [2:26:22<00:00, 11.39it/s]\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "torch_counter"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "gVER_I7TbCEU",
        "outputId": "d497d3da-746b-4306-be4b-bffae39412b5"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "Counter({(0, 0, 0): 1822396240,\n",
              "         (255, 255, 255): 1823964221,\n",
              "         (190, 190, 190): 2889984,\n",
              "         (128, 128, 128): 739457791,\n",
              "         (198, 198, 198): 543920,\n",
              "         (125, 125, 125): 3190558,\n",
              "         (127, 127, 127): 56406167,\n",
              "         (110, 110, 110): 195534762,\n",
              "         (240, 240, 240): 214896165,\n",
              "         (145, 145, 145): 19869696,\n",
              "         (187, 187, 187): 3145868,\n",
              "         (243, 243, 243): 30150259,\n",
              "         (220, 220, 220): 9935125,\n",
              "         (130, 130, 130): 297840,\n",
              "         (101, 101, 101): 173303,\n",
              "         (119, 119, 119): 63610,\n",
              "         (137, 137, 137): 63646,\n",
              "         (155, 155, 155): 171799,\n",
              "         (254, 254, 254): 18916352,\n",
              "         (98, 98, 98): 17136000,\n",
              "         (146, 146, 146): 50787,\n",
              "         (163, 163, 163): 48868,\n",
              "         (92, 92, 92): 48847,\n",
              "         (209, 209, 209): 9077003,\n",
              "         (96, 96, 96): 10135552,\n",
              "         (132, 132, 132): 2797246,\n",
              "         (83, 83, 83): 481,\n",
              "         (172, 172, 172): 484,\n",
              "         (188, 188, 188): 2778080,\n",
              "         (15, 15, 15): 20973568,\n",
              "         (45, 45, 45): 9533440,\n",
              "         (80, 80, 80): 457488,\n",
              "         (69, 69, 69): 6354,\n",
              "         (201, 201, 201): 66800,\n",
              "         (216, 216, 216): 63987,\n",
              "         (197, 197, 197): 172800,\n",
              "         (199, 199, 199): 376240,\n",
              "         (131, 131, 131): 186720,\n",
              "         (129, 129, 129): 326320,\n",
              "         (133, 133, 133): 33098,\n",
              "         (219, 219, 219): 32029,\n",
              "         (204, 204, 204): 45520,\n",
              "         (205, 205, 205): 38664,\n",
              "         (203, 203, 203): 46720,\n",
              "         (225, 225, 225): 13481,\n",
              "         (215, 215, 215): 61520,\n",
              "         (223, 223, 223): 31120,\n",
              "         (217, 217, 217): 41231,\n",
              "         (126, 126, 126): 222572,\n",
              "         (123, 123, 123): 27703,\n",
              "         (210, 210, 210): 73329,\n",
              "         (124, 124, 124): 33390,\n",
              "         (196, 196, 196): 12800,\n",
              "         (229, 229, 229): 8528,\n",
              "         (213, 213, 213): 36772,\n",
              "         (206, 206, 206): 38480,\n",
              "         (212, 212, 212): 33043,\n",
              "         (135, 135, 135): 8455,\n",
              "         (207, 207, 207): 36960,\n",
              "         (221, 221, 221): 33533,\n",
              "         (211, 211, 211): 19600,\n",
              "         (200, 200, 200): 146080,\n",
              "         (134, 134, 134): 18174,\n",
              "         (226, 226, 226): 19331,\n",
              "         (227, 227, 227): 3200,\n",
              "         (222, 222, 222): 36683,\n",
              "         (208, 208, 208): 26435,\n",
              "         (232, 232, 232): 5172,\n",
              "         (228, 228, 228): 3015,\n",
              "         (214, 214, 214): 27040,\n",
              "         (122, 122, 122): 8144,\n",
              "         (202, 202, 202): 29840,\n",
              "         (218, 218, 218): 12480,\n",
              "         (234, 234, 234): 4593,\n",
              "         (233, 233, 233): 4628,\n",
              "         (230, 230, 230): 5082,\n",
              "         (224, 224, 224): 17184})"
            ]
          },
          "metadata": {},
          "execution_count": 18
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "gradient = gradient_value(collections.Counter(torch_counter), 112 ** 2)\n",
        "b = [gradient[i:i + col] for i in range(0, len(gradient), col)]\n",
        "b = list(zip(*b))\n",
        "Image.fromarray(np.array(b)).convert('L')"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 177
        },
        "id": "I9wh5bgF8nhu",
        "outputId": "d92555fb-c0bd-4db4-d275-c428397cfb76"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<PIL.Image.Image image mode=L size=2000x200>"
            ],
            "image/png": "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\n"
          },
          "metadata": {},
          "execution_count": 19
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [],
      "metadata": {
        "id": "KF6AEmfW8s6H"
      },
      "execution_count": null,
      "outputs": []
    }
  ]
}